From 5c3a726041a44ab20c6177721cffb974ea576d77 Mon Sep 17 00:00:00 2001
From: LIghtJUNction <lightjunction.me@gmail.com>
Date: Mon, 1 Jun 2026 04:26:35 +0800
Subject: [PATCH 1/4] chore: migrate skills to agents layout

---
 .../skills}/_template-episodic.md             |  6 +-
 .../skills}/_template-procedural.md           |  6 +-
 .../skills}/_template-semantic.md             |  6 +-
 {skills => .agents/skills}/_template.md       |  2 +-
 .../skills}/clinical-medicine/.gitkeep        |  0
 ...enchmark-prompt-must-strip-gold-answers.md |  0
 ...riage-before-cross-round-interpretation.md |  0
 ...n-incident-source-links-are-unreachable.md |  0
 ...ions-around-talent-led-option-formation.md |  0
 ...y-occupies-the-measurement-contribution.md |  0
 ...en-open-question-banks-are-inaccessible.md |  0
 ...ctor-signals-shared-benchmark-structure.md |  0
 ...ication-when-within-firm-signal-is-thin.md |  0
 ...ion-for-exploratory-incident-taxonomies.md |  0
 ...pectacular-but-brittle-proposal-framing.md |  0
 ...equires-an-adversarial-intent-dimension.md |  0
 ...xterity-product-is-combined-not-balance.md |  0
 ...chmark-interpretation-for-management-qa.md |  0
 ...on-theory-is-not-finance-binary-options.md |  0
 ...ization-into-management-informatization.md |  0
 ...dogeneity-tests-do-not-prove-exogeneity.md |  0
 ...refers-control-function-over-naive-2sls.md |  0
 ...librated-extension-not-paradigm-pioneer.md |  0
 ...by-subdimension-not-by-total-item-count.md |  0
 ...ilures-can-masquerade-as-knowledge-gaps.md |  0
 ...-memory-by-dropping-non-backprop-models.md |  0
 ...ilure--token-level-grpo-mask-invariance.md |  0
 ...cuda-memory-gap-fragmentation-diagnosis.md |  0
 ...-external-model-outputs-before-training.md |  0
 ...-grpo-numerical-stability-kl-divergence.md |  0
 .../tie--bnpo-vs-dapo-loss-scaling-choice.md  |  0
 ...gradient-checkpointing-target-selection.md |  0
 ...-llm-benchmarking-regression-heuristics.md |  0
 ...oducibility-seed-limitations-in-pytorch.md |  0
 ...kenizer-invariance-during-model-updates.md |  0
 ...frontier-verifiability-gates-automation.md |  0
 ...rong-granularity-for-token-level-priors.md |  0
 ...t-cutoff-model-partial-recognition-trap.md |  0
 ...on-as-skill-signal-in-conversation-logs.md |  0
 ...ructure-layer-question-before-injecting.md |  0
 ...luate-research-skills-without-benchmark.md |  0
 ...ctory-is-higher-signal-than-final-paper.md |  0
 ...l--conference-topic-from-x-to-y-framing.md |  0
 ...an-category-implementation-start-narrow.md |  0
 ...arch-skill-format-four-section-template.md |  0
 ...capacity-is-binding-in-ai-research-2026.md |  0
 ...ling-condition-not-standalone-direction.md |  0
 ...research-direction-closed-loop-ordering.md |  0
 ...earch-skill-definition-three-conditions.md |  0
 ...-skill-library-beats-deep-autogenerated.md |  0
 ...age-inverse-proportional-to-skill-value.md |  0
 ...skillsbench-negative-transfer-mechanism.md |  0
 ...ttokens-ieee754-single-token-arithmetic.md |  0
 ...m-separating-computation-from-retrieval.md |  0
 ...n-h-hybrid-92-percent-mamba-replacement.md |  0
 ...scientist-nature-publication-march-2026.md |  0
 ...ator-quality-dominates-executor-quality.md |  0
 ...ft-hierarchical-skill-composition-fails.md |  0
 ...dgment-matters-more-than-skill-creation.md |  0
 ...bench-human-vs-self-generated-skill-gap.md |  0
 ...kenization-right-to-left-for-arithmetic.md |  0
 ...ntier--ttrl-test-time-rl-without-labels.md |  0
 ...val-continuous-number-embedding-scaling.md |  0
 ...noise-floor-3-to-5-signal-turns-per-200.md |  0
 ...rational-failure-modes-to-guard-against.md |  0
 ...mpty-responses-from-chat-format-prompts.md |  0
 ...omparison-table-averaging-formula-audit.md |  0
 ...-rarely-has-flawed-process-in-rl-models.md |  0
 ...nk-ordering-guarantee-dual-advantage-rl.md |  0
 ...-habit-emerges-from-process-supervision.md |  0
 ...-framed-as-the-brain-written-as-formula.md |  0
 ...comes-from-defining-the-metric-boundary.md |  0
 ...s-need-scaling-and-input-contract-tests.md |  0
 ...-initial-solve-versus-post-optimization.md |  0
 ...e-overfit-into-public-scientific-claims.md |  0
 ...ation-needs-stability-not-just-best-run.md |  0
 ...splits-must-avoid-design-family-leakage.md |  0
 ...r-ablation-for-model-compression-claims.md |  0
 ...-claims-require-label-budget-accounting.md |  0
 ...-needs-outcome-credit-not-only-token-kl.md |  0
 ...a-hidden-confounder-in-agent-benchmarks.md |  0
 ...riven-placement-must-end-in-sta-metrics.md |  0
 ...es-need-path-topology-not-just-geometry.md |  0
 ...-should-be-kept-as-reviewer-risk-memory.md |  0
 ...-clustered-into-experiment-design-fixes.md |  0
 ...ters-need-a-single-philosophical-thread.md |  0
 ...should-trim-detours-not-dilute-concepts.md |  0
 ...dients-require-bias-and-variance-audits.md |  0
 ...put-drift-not-only-retained-token-score.md |  0
 ...ons-need-solver-interface-specification.md |  0
 ...hould-start-from-trajectory-probability.md |  0
 ...red-when-contrasting-learning-paradigms.md |  0
 ...acement-needs-gradient-usefulness-check.md |  0
 ...vidence-for-learning-regime-comparisons.md |  0
 ...aims-need-proxy-to-signoff-causal-chain.md |  0
 ...be-history-of-ideas-not-event-chronicle.md |  0
 ...tic-should-be-positioned-conservatively.md |  0
 ...-speculative-reasoning-not-a-theme-word.md |  0
 ...-as-relative-advantage-without-a-critic.md |  0
 ...-computability-origin-not-name-dropping.md |  0
 ...resented-as-three-lenses-not-one-slogan.md |  0
 ...ppo-by-supervision-source-and-objective.md |  0
 ...ai-origin-arc-should-be-mechanism-first.md |  0
 ...ai-history-needs-modern-llm-return-path.md |  0
 ...ant-than-task-count-in-agent-benchmarks.md |  0
 ...mportance-is-not-enough-for-kv-eviction.md |  0
 ...-solution-batches-not-marginal-averages.md |  0
 ...-scores-do-not-imply-a-hidden-threshold.md |  0
 ...ning-is-not-generic-model-collaboration.md |  0
 ...ameter-is-not-equivalent-to-high-or-off.md |  0
 ...n-use-olympiad-style-continuous-scoring.md |  0
 ...s-reliability-not-single-run-competence.md |  0
 ...-space-projection-for-kv-cache-eviction.md |  0
 ...isagreement-in-cross-version-validation.md |  0
 ...iling-full-system-numerical-equivalence.md |  0
 ...response-removal-before-formal-analysis.md |  0
 ...ervention-claims-in-stress-level-thesis.md |  0
 ...t-after-changing-h3-to-joint-prediction.md |  0
 ...mpling-wording-must-match-actual-design.md |  0
 ...sting-method-structure-when-fixing-math.md |  0
 ...tive-and-negative-methodological-claims.md |  0
 ...ct-prompts-to-the-single-swapped-object.md |  0
 ...claims-as-evidence-of-semantic-recovery.md |  0
 ...mple-mining-from-near-miss-object-pairs.md |  0
 ...sms-are-scientific-vs-merely-rhetorical.md |  0
 ...cal-model-output-not-observed-responses.md |  0
 ...is-a-separate-target-not-human-accuracy.md |  0
 ...l-similarity-not-label-level-similarity.md |  0
 ...n-rl-style-alignment-in-this-evaluation.md |  0
 ...e-variance-without-moving-the-mean-much.md |  0
 ...-even-when-severe-errors-still-dominate.md |  0
 ...vidence-grounding-in-experiment-reports.md |  0
 ...re--research-code-release-blocker-audit.md |  0
 ...n-readiness-checks-for-anonymous-papers.md |  0
 ...oducible-job-scripts-for-ml-experiments.md |  0
 ...e--evidence-backed-work-timeline-blocks.md |  0
 ...chronological-experiment-result-logging.md |  0
 ...-preamble-isolation-for-paper-templates.md |  0
 ...obes-for-retrospective-paper-evaluation.md |  0
 ...ological-rigor-in-paper-diff-assessment.md |  0
 ...view-found-hidden-data-manuscript-drift.md |  0
 ...ve-practices-to-reject-in-paper-writing.md |  0
 ...s-for-avoiding-llm-writing-antipatterns.md |  0
 ...rate-regression-after-feature-reduction.md |  0
 ...ent-paper-algorithm-not-generic-concept.md |  0
 ...son-axis-when-caught-in-false-dichotomy.md |  0
 ...tal-credibility-in-cherry-picked-fields.md |  0
 ...-ontology-free-multi-agent-coordination.md |  0
 ...ory-extraction-needs-post-pass-curation.md |  0
 ...e-signals-a-research-data-turning-point.md |  0
 ...er-chat-diffing-reveals-tacit-decisions.md |  0
 ...with-escape-hatch-for-research-ontology.md |  0
 ...ry-types-for-scientific-knowhow-capture.md |  0
 ...harness-as-research-workflow-primitives.md |  0
 ...ark-environment-for-ai-agent-scientists.md |  0
 ...-skills-registry-for-ai-research-agents.md |  0
 ...c--researchskills-skill-memory-taxonomy.md |  0
 ...s-fluency-as-scientific-model-objective.md |  0
 ...-topic-selection-after-field-maturation.md |  0
 ...-into-build-integrate-optimize-sequence.md |  0
 .../skills}/management/.gitkeep               |  0
 .../skills}/management/careers/.gitkeep       |  0
 .../.gitkeep                                  |  0
 .../management/conflict-management/.gitkeep   |  0
 .../critical-management-studies/.gitkeep      |  0
 .../diversity-equity-and-inclusion/.gitkeep   |  0
 .../management/entrepreneurship/.gitkeep      |  0
 .../health-care-management/.gitkeep           |  0
 .../management/human-resources/.gitkeep       |  0
 .../international-management/.gitkeep         |  0
 .../management/management-consulting/.gitkeep |  0
 .../.gitkeep                                  |  0
 .../management/management-history/.gitkeep    |  0
 .../.gitkeep                                  |  0
 .../.gitkeep                                  |  0
 .../.gitkeep                                  |  0
 .../.gitkeep                                  |  0
 .../.gitkeep                                  |  0
 .../organizational-behavior/.gitkeep          |  0
 .../.gitkeep                                  |  0
 .../.gitkeep                                  |  0
 .../management/public-and-nonprofit/.gitkeep  |  0
 .../management/research-methods/.gitkeep      |  0
 .../social-issues-in-management/.gitkeep      |  0
 .../management/strategic-management/.gitkeep  |  0
 .../.gitkeep                                  |  0
 .../.gitkeep                                  |  0
 ...ce-speedup-via-weierstrass-construction.md |  0
 ...rnoon-psn-depression-vs-shs-enhancement.md |  0
 ...ous--phs-subtropical-degradation-28-32n.md |  0
 ...n-violates-standard-water-balance-check.md |  0
 ...ty-as-minimum-bar-for-scientific-claims.md |  0
 ...--legacy-phs-v3-conus-domain-invalidity.md |  0
 ...hs-vpsis-physical-clamp-at-minus-9000mm.md |  0
 ...ias-before-helmholtz-divergence-removal.md |  0
 ...--helmholtz-projection-nyquist-residual.md |  0
 ...public--laps-mhd-output-float-precision.md |  0
 ...nvariance-in-sphere-packing-simulations.md |  0
 ...model-architecture-in-rl-packing-policy.md |  0
 ...g-regime-in-polydisperse-sphere-packing.md |  0
 ...nce-in-reinforce-based-packing-policies.md |  0
 ...ates-analytically-in-sequential-packing.md |  0
 ...s-training-failure-in-rl-sphere-packing.md |  0
 ...the-manuscript-not-to-the-rubric-itself.md |  0
 ...d-frontier-gap-scan-for-review-articles.md |  0
 ...on-interactive-webapp-as-differentiator.md |  0
 ...n--hysteresis-scanning-curve-validation.md |  0
 ...al-model-mismatch-via-seasonal-dynamics.md |  0
 ...olution-vs-rock-moisture-scale-mismatch.md |  0
 ...logy-model-integration-of-rock-moisture.md |  0
 ...-failure--llm-agent-fortran-refactoring.md |  0
 ...-surface-products-for-coastal-turbidity.md |  0
 ...nnamed-datasets-require-named-citations.md |  0
 ...w-upscaling-uncertainty-into-likelihood.md |  0
 ...g-overconfidence-induced-research-stall.md |  0
 ...r-tool-optimization-in-research-systems.md |  0
 ...position-under-vague-problem-definition.md |  0
 ...-cutting-validation-four-variable-types.md |  0
 ...-fail--dual-porosity-convergence-tuning.md |  0
 ...e-kalman-filter-smap-grace-localization.md |  0
 ...olate-rock-vs-groundwater-contributions.md |  0
 ...ependent-domain-path-dependent-psi-stem.md |  0
 ...iability-success-criteria-rock-moisture.md |  0
 ...ter-error-vs-structural-error-in-models.md |  0
 ...in-concepts-for-hydrological-innovation.md |  0
 ...james-vs-ai-venue-for-ai-agents-for-esm.md |  0
 ...er-finetuning-for-foundation-embeddings.md |  0
 ...tterns-over-pointwise-fit-in-geoscience.md |  0
 ...eam-for-hydraulic-parameter-calibration.md |  0
 ...bration-to-manage-parameter-compounding.md |  0
 ...-estimation-is-data-assimilation-naming.md |  0
 ...s-not-a-side-note-in-modern-lsm-reviews.md |  0
 ...need-benchmark-and-mechanism-qualifiers.md |  0
 ...-sapwood-50-percent-daily-transpiration.md |  0
 ...is-a-design-choice-not-a-physical-limit.md |  0
 ...n--noah-mp-wood-moisture-initialization.md |  0
 ...-dark-reactions-decoupling-under-stress.md |  0
 ...ve-one-out-cross-validation-distinction.md |  0
 ...figures-correct-semantic-interpretation.md |  0
 ...alphaearth-10m-64band-annual-embeddings.md |  0
 ...based-mass-balance-lower-bound-sbedrock.md |  0
 ...nonymous-peer-review-formatting-finesst.md |  0
 ...rontier--mcp-physics-models-novelty-gap.md |  0
 ...tection-radon-transform-vs-cnn-pipeline.md |  0
 ...-rock-moisture-storage-capacity-scaling.md |  0
 ...n-genuchten-hysteresis-stem-water-curve.md |  0
 ...ga-tonga-lamb-wave-two-mode-propagation.md |  0
 ...-meteor-radar-double-gaussian-dual-peak.md |  0
 ...mp-dual-porosity-bedrock-extension-spec.md |  0
 ...-noah-mp-ncar-hpc-allocation-cpu-budget.md |  0
 ...lith-thickness-dataset-for-noahmp-depth.md |  0
 ...-phase-1-prior-work-line-counts-fortran.md |  0
 ...eckpoint-reuse-to-skip-s1-re-simulation.md |  0
 ...sue-limit-cycle-is-pacing-site-specific.md |  0
 ...ng-sites-are-endocardial-not-epicardial.md |  0
 ...ce-first-visualization-over-raw-spectra.md |  0
 ...tellite-relative-motion-reference-frame.md |  0
 ...e-advection-dominates-spacecraft-motion.md |  0
 ...s-sampling-artifact-of-moving-structure.md |  0
 ...tour-is-not-the-flux-rope-cross-section.md |  0
 ...ation-proxy-not-a-true-3d-magnetic-axis.md |  0
 ...ula-full-vs-simplified-form-discrepancy.md |  0
 ...ated-conductance-not-local-conductivity.md |  0
 ...ection-in-multi-instrument-reproduction.md |  0
 ...urce-model-for-satellite-plotting-tools.md |  0
 ...on--trvae-unpaired-condition-generation.md |  0
 ...ilure--slaf-large-scale-anndata-merging.md |  0
 ...ion-invariant-program-recovery-protocol.md |  0
 ...--graph-guided-spatial-programs-need-uq.md |  0
 ...n-corrupt-ranking-evaluation-indirectly.md |  0
 ...split-diagnosis-with-test-only-entities.md |  0
 ...nking-metrics-in-combination-prediction.md |  0
 ...-when-features-are-random-or-simplified.md |  0
 ...der-generalization-as-order-composition.md |  0
 ...ss-edge-source-must-follow-split-regime.md |  0
 ...mposed-imitation-loss-for-order-experts.md |  0
 ...ion-for-combination-entity-conditioning.md |  0
 ...raphic-order-decomposition-for-unseen-k.md |  0
 ...sitive-versus-multi-positive-mrr-choice.md |  0
 ...or-fail--gp-kernel-psd-and-jitter-check.md |  0
 ...s-validation-by-column-masking-in-rasch.md |  0
 ...li-logprob-loss-interpretation-in-rasch.md |  0
 ...auc-computation-with-masked-matrix-data.md |  0
 ...to-sgmm-when-spatial-clustering-matters.md |  0
 ...ublic--psmil-bmle-smle-partial-labeling.md |  0
 ...m-spatially-varying-mixture-proportions.md |  0
 AGENT.md                                      | 13 +++
 readme.md => README.md                        | 60 +++++++++++-
 readme_zh.md                                  | 81 +++++++++++++++-
 .../commands/CONVERT-SKILL.md                 | 14 +--
 .../commands/researchskills-convert.md        | 14 +--
 .../docs/codex-migration-plan.md              |  2 +-
 .../templates/skill-template.md               |  2 +-
 utils/CONTRIBUTING.md                         | 26 ++---
 utils/SKILL_SCHEMA.md                         | 94 ++++++++++++-------
 utils/scripts/generate-taxonomy.js            |  8 +-
 utils/tools/validate.py                       | 78 +++++++++------
 297 files changed, 299 insertions(+), 113 deletions(-)
 rename {skills => .agents/skills}/_template-episodic.md (89%)
 rename {skills => .agents/skills}/_template-procedural.md (89%)
 rename {skills => .agents/skills}/_template-semantic.md (88%)
 rename {skills => .agents/skills}/_template.md (96%)
 rename {skills => .agents/skills}/clinical-medicine/.gitkeep (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--benchmark-prompt-must-strip-gold-answers.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--noise-first-item-triage-before-cross-round-interpretation.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--provenance-validation-when-incident-source-links-are-unreachable.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-legacy-digital-options-around-talent-led-option-formation.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-novelty-claims-when-prior-literature-already-occupies-the-measurement-contribution.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--theory-grounded-supplementation-when-open-question-banks-are-inaccessible.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--convergent-wrong-distractor-signals-shared-benchmark-structure.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--fixed-effects-can-destroy-identification-when-within-firm-signal-is-thin.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--no-single-k-cluster-validation-for-exploratory-incident-taxonomies.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/episodic/failure--prefer-mechanism-grounded-metaphors-over-spectacular-but-brittle-proposal-framing.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ai-incident-theory-requires-an-adversarial-intent-dimension.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ambidexterity-product-is-combined-not-balance.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--contamination-aware-benchmark-interpretation-for-management-qa.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--digital-option-theory-is-not-finance-binary-options.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--do-not-collapse-governance-digitalization-into-management-informatization.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--non-significant-endogeneity-tests-do-not-prove-exogeneity.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--nonlinear-endogeneity-prefers-control-function-over-naive-2sls.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--novelty-in-agentic-is-should-be-framed-as-calibrated-extension-not-paradigm-pioneer.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--rebalance-benchmark-datasets-by-subdimension-not-by-total-item-count.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/clarence-zhewang/semantic/frontier--tokenizer-level-name-failures-can-masquerade-as-knowledge-gaps.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/episodic/adaptation--freeing-memory-by-dropping-non-backprop-models.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/procedural/constraint-failure--token-level-grpo-mask-invariance.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--cuda-memory-gap-fragmentation-diagnosis.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--detaching-external-model-outputs-before-training.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--grpo-numerical-stability-kl-divergence.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/procedural/tie--bnpo-vs-dapo-loss-scaling-choice.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/procedural/tie--gradient-checkpointing-target-selection.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/procedural/tie--llm-benchmarking-regression-heuristics.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/semantic/correction--reproducibility-seed-limitations-in-pytorch.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ergan-shang/semantic/correction--tokenizer-invariance-during-model-updates.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--jagged-frontier-verifiability-gates-automation.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--kantian-categories-wrong-granularity-for-token-level-priors.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--post-cutoff-model-partial-recognition-trap.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--direction-negation-as-skill-signal-in-conversation-logs.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--embedding-structure-layer-question-before-injecting.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--evaluate-research-skills-without-benchmark.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--paper-trajectory-is-higher-signal-than-final-paper.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--conference-topic-from-x-to-y-framing.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--kantian-category-implementation-start-narrow.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--research-skill-format-four-section-template.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--direction-not-capacity-is-binding-in-ai-research-2026.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--mechanistic-interpretability-is-enabling-condition-not-standalone-direction.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-direction-closed-loop-ordering.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-skill-definition-three-conditions.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--shallow-validated-skill-library-beats-deep-autogenerated.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-domain-coverage-inverse-proportional-to-skill-value.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-negative-transfer-mechanism.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--bittokens-ieee754-single-token-arithmetic.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--deepseek-engram-separating-computation-from-retrieval.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--nemotron-h-hybrid-92-percent-mamba-replacement.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--sakana-ai-scientist-nature-publication-march-2026.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-creator-quality-dominates-executor-quality.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-hierarchical-skill-composition-fails.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-skill-judgment-matters-more-than-skill-creation.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillsbench-human-vs-self-generated-skill-gap.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--tokenization-right-to-left-for-arithmetic.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--ttrl-test-time-rl-without-labels.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--xval-continuous-number-embedding-scaling.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/hhhhhejia/semantic/non-public--conversation-noise-floor-3-to-5-signal-turns-per-200.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/ktwu01/procedural/tie--ai-agent-discipline-operational-failure-modes-to-guard-against.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--base-model-empty-responses-from-chat-format-prompts.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--comparison-table-averaging-formula-audit.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--correct-answer-rarely-has-flawed-process-in-rl-models.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--rank-ordering-guarantee-dual-advantage-rl.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/tanzelin430/semantic/non-public--verification-habit-emerges-from-process-supervision.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--mcculloch-pitts-is-best-framed-as-the-brain-written-as-formula.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--rebuttal-efficiency-comes-from-defining-the-metric-boundary.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--arbitrary-size-gnn-claims-need-scaling-and-input-contract-tests.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--inference-time-must-define-initial-solve-versus-post-optimization.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--internal-benchmark-results-should-not-be-overfit-into-public-scientific-claims.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--llm-agent-evaluation-needs-stability-not-just-best-run.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--placement-benchmark-splits-must-avoid-design-family-leakage.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--same-parameter-ablation-for-model-compression-claims.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--semi-supervised-claims-require-label-budget-accounting.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--sequence-level-agent-training-needs-outcome-credit-not-only-token-kl.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--thinking-budget-is-a-hidden-confounder-in-agent-benchmarks.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-driven-placement-must-end-in-sta-metrics.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-surrogates-need-path-topology-not-just-geometry.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--dead-end-experiments-should-be-kept-as-reviewer-risk-memory.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--reviewer-concerns-should-be-clustered-into-experiment-design-fixes.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-chapters-need-a-single-philosophical-thread.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-popularization-should-trim-detours-not-dilute-concepts.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--differentiable-ilp-gradients-require-bias-and-variance-audits.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--kv-eviction-should-be-evaluated-by-output-drift-not-only-retained-token-score.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--neural-milp-solutions-need-solver-interface-specification.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--policy-gradient-explanations-should-start-from-trajectory-probability.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--related-work-baselines-are-required-when-contrasting-learning-paradigms.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--timing-driven-global-placement-needs-gradient-usefulness-check.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--training-curve-evidence-for-learning-regime-comparisons.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-eda-paper-claims-need-proxy-to-signoff-causal-chain.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-history-book-positioning-should-be-history-of-ideas-not-event-chronicle.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--benchmark-release-from-internal-diagnostic-should-be-positioned-conservatively.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--entropy-is-a-routing-signal-for-speculative-reasoning-not-a-theme-word.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--grpo-should-be-framed-as-relative-advantage-without-a-critic.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--hilbert-godel-turing-arc-should-be-used-as-computability-origin-not-name-dropping.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--information-control-systems-should-be-presented-as-three-lenses-not-one-slogan.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--on-policy-distillation-differs-from-ppo-by-supervision-source-and-objective.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--pre-dartmouth-ai-origin-arc-should-be-mechanism-first.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--scholarly-ai-history-needs-modern-llm-return-path.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/procedural/tie--task-success-criteria-are-more-important-than-task-count-in-agent-benchmarks.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/correction--attention-weight-importance-is-not-enough-for-kv-eviction.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/correction--diffusion-for-milp-should-generate-solution-batches-not-marginal-averages.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/correction--llm-judge-binary-scores-do-not-imply-a-hidden-threshold.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/correction--multi-model-speculative-reasoning-is-not-generic-model-collaboration.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/correction--omitted-thinking-parameter-is-not-equivalent-to-high-or-off.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/correction--proof-benchmarks-can-use-olympiad-style-continuous-scoring.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--pass-three-measures-reliability-not-single-run-competence.md (100%)
 rename {skills => .agents/skills}/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--value-space-projection-for-kv-cache-eviction.md (100%)
 rename {skills => .agents/skills}/computer-science/computational-engineering-finance-and-science/ktwu01/episodic/anomalous--incomplete-archived-simulations-can-masquerade-as-scientific-disagreement-in-cross-version-validation.md (100%)
 rename {skills => .agents/skills}/computer-science/computational-engineering-finance-and-science/ktwu01/semantic/correction--matching-hydrology-options-can-preserve-a-port-while-still-failing-full-system-numerical-equivalence.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/episodic/adaptation--invalid-survey-response-removal-before-formal-analysis.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--avoid-causal-or-intervention-claims-in-stress-level-thesis.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--dissertation-hypothesis-alignment-after-changing-h3-to-joint-prediction.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--sampling-wording-must-match-actual-design.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--preserve-existing-method-structure-when-fixing-math.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--reviewer-summary-should-separate-positive-and-negative-methodological-claims.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reduce-multi-object-prompts-to-the-single-swapped-object.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reframe-failed-ranking-claims-as-evidence-of-semantic-recovery.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--soft-hallucination-example-mining-from-near-miss-object-pairs.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/failure--reviewer-pushback-can-reveal-which-criticisms-are-scientific-vs-merely-rhetorical.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--balanced-hard-baseline-comes-from-the-hypothetical-model-output-not-observed-responses.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--human-score-is-a-separate-target-not-human-accuracy.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--object-level-similarity-not-label-level-similarity.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--alignment-tuning-softens-hallucination-severity-more-reliably-than-rl-style-alignment-in-this-evaluation.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--multi-view-human-alignment-can-increase-variance-without-moving-the-mean-much.md (100%)
 rename {skills => .agents/skills}/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--stronger-models-show-fewer-severe-hallucinations-even-when-severe-errors-still-dominate.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--evidence-grounding-in-experiment-reports.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--research-code-release-blocker-audit.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--submission-readiness-checks-for-anonymous-papers.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/a-green-hand-jack/procedural/no-change--reproducible-job-scripts-for-ml-experiments.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/a-green-hand-jack/procedural/tie--evidence-backed-work-timeline-blocks.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/a-green-hand-jack/procedural/tie--reverse-chronological-experiment-result-logging.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/a-green-hand-jack/semantic/correction--venue-preamble-isolation-for-paper-templates.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/ktwu01/episodic/adaptation--commit-log-probes-for-retrospective-paper-evaluation.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/ktwu01/episodic/failure--auto-review-narratives-can-overstate-methodological-rigor-in-paper-diff-assessment.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/ktwu01/episodic/failure--benchmark-paper-review-found-hidden-data-manuscript-drift.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/ktwu01/procedural/tie--research-integrity-antipatterns-deceptive-practices-to-reject-in-paper-writing.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/ktwu01/procedural/tie--scientific-prose-style-rules-for-avoiding-llm-writing-antipatterns.md (100%)
 rename {skills => .agents/skills}/computer-science/machine-learning/zhuofli/episodic/failure--diagnosing-degenerate-regression-after-feature-reduction.md (100%)
 rename {skills => .agents/skills}/computer-science/multiagent-systems/xisen-w/episodic/failure--proof-of-concept-must-implement-paper-algorithm-not-generic-concept.md (100%)
 rename {skills => .agents/skills}/computer-science/multiagent-systems/xisen-w/procedural/constraint-failure--reframe-comparison-axis-when-caught-in-false-dichotomy.md (100%)
 rename {skills => .agents/skills}/computer-science/multiagent-systems/xisen-w/procedural/operator-fail--explicit-regime-boundaries-increase-rebuttal-credibility-in-cherry-picked-fields.md (100%)
 rename {skills => .agents/skills}/computer-science/multiagent-systems/xisen-w/semantic/frontier--llms-as-universal-translators-enable-ontology-free-multi-agent-coordination.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/anon-b34ee36c/episodic/adaptation--research-memory-extraction-needs-post-pass-curation.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/anon-b34ee36c/episodic/anomalous--unexpected-template-mixture-in-a-single-site-archive-signals-a-research-data-turning-point.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/anon-b34ee36c/procedural/tie--claim-level-paper-chat-diffing-reveals-tacit-decisions.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/anon-b34ee36c/procedural/tie--predefined-action-vocabulary-with-escape-hatch-for-research-ontology.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/anon-b34ee36c/semantic/correction--three-memory-types-for-scientific-knowhow-capture.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/hhhhhejia/semantic/non-public--agent-harness-as-research-workflow-primitives.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/hhhhhejia/semantic/non-public--benchmark-environment-for-ai-agent-scientists.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/hhhhhejia/semantic/non-public--human-authored-skills-registry-for-ai-research-agents.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/hhhhhejia/semantic/non-public--researchskills-skill-memory-taxonomy.md (100%)
 rename {skills => .agents/skills}/computer-science/software-engineering/hhhhhejia/semantic/non-public--skills-fluency-as-scientific-model-objective.md (100%)
 rename {skills => .agents/skills}/economics/general-economics/clarence-zhewang/semantic/frontier--data-factor-topic-selection-after-field-maturation.md (100%)
 rename {skills => .agents/skills}/economics/theoretical-economics/clarence-zhewang/episodic/adaptation--transform-parallel-studies-into-build-integrate-optimize-sequence.md (100%)
 rename {skills => .agents/skills}/management/.gitkeep (100%)
 rename {skills => .agents/skills}/management/careers/.gitkeep (100%)
 rename {skills => .agents/skills}/management/communication-digital-technology-and-organization/.gitkeep (100%)
 rename {skills => .agents/skills}/management/conflict-management/.gitkeep (100%)
 rename {skills => .agents/skills}/management/critical-management-studies/.gitkeep (100%)
 rename {skills => .agents/skills}/management/diversity-equity-and-inclusion/.gitkeep (100%)
 rename {skills => .agents/skills}/management/entrepreneurship/.gitkeep (100%)
 rename {skills => .agents/skills}/management/health-care-management/.gitkeep (100%)
 rename {skills => .agents/skills}/management/human-resources/.gitkeep (100%)
 rename {skills => .agents/skills}/management/international-management/.gitkeep (100%)
 rename {skills => .agents/skills}/management/management-consulting/.gitkeep (100%)
 rename {skills => .agents/skills}/management/management-education-and-development/.gitkeep (100%)
 rename {skills => .agents/skills}/management/management-history/.gitkeep (100%)
 rename {skills => .agents/skills}/management/management-spirituality-and-religion/.gitkeep (100%)
 rename {skills => .agents/skills}/management/managerial-and-organizational-cognition/.gitkeep (100%)
 rename {skills => .agents/skills}/management/operations-and-supply-chain-management/.gitkeep (100%)
 rename {skills => .agents/skills}/management/organization-and-management-theory/.gitkeep (100%)
 rename {skills => .agents/skills}/management/organization-development-and-change/.gitkeep (100%)
 rename {skills => .agents/skills}/management/organizational-behavior/.gitkeep (100%)
 rename {skills => .agents/skills}/management/organizational-neuroscience-and-biology/.gitkeep (100%)
 rename {skills => .agents/skills}/management/organizations-and-the-natural-environment/.gitkeep (100%)
 rename {skills => .agents/skills}/management/public-and-nonprofit/.gitkeep (100%)
 rename {skills => .agents/skills}/management/research-methods/.gitkeep (100%)
 rename {skills => .agents/skills}/management/social-issues-in-management/.gitkeep (100%)
 rename {skills => .agents/skills}/management/strategic-management/.gitkeep (100%)
 rename {skills => .agents/skills}/management/strategizing-activities-and-practices/.gitkeep (100%)
 rename {skills => .agents/skills}/management/technology-and-innovation-management/.gitkeep (100%)
 rename {skills => .agents/skills}/mathematics/approximation-theory/pillow0705/semantic/frontier--polynomial-approximation-of-neural-networks-achieves-16x-inference-speedup-via-weierstrass-construction.md (100%)
 rename {skills => .agents/skills}/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-afternoon-psn-depression-vs-shs-enhancement.md (100%)
 rename {skills => .agents/skills}/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-subtropical-degradation-28-32n.md (100%)
 rename {skills => .agents/skills}/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-two-pathway-transpiration-violates-standard-water-balance-check.md (100%)
 rename {skills => .agents/skills}/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/failure--falsifiability-as-minimum-bar-for-scientific-claims.md (100%)
 rename {skills => .agents/skills}/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--legacy-phs-v3-conus-domain-invalidity.md (100%)
 rename {skills => .agents/skills}/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--phs-vpsis-physical-clamp-at-minus-9000mm.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/huangzesen/procedural/constraint-failure--dealias-before-helmholtz-divergence-removal.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/huangzesen/semantic/correction--helmholtz-projection-nyquist-residual.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/huangzesen/semantic/non-public--laps-mhd-output-float-precision.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/pillow0705/episodic/anomalous--absolute-collision-tolerance-breaks-scale-invariance-in-sphere-packing-simulations.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/pillow0705/procedural/tie--particle-size-distribution-shape-dominates-model-architecture-in-rl-packing-policy.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/pillow0705/semantic/frontier--embedding-ratio-threshold-determines-packing-regime-in-polydisperse-sphere-packing.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/pillow0705/semantic/frontier--extended-training-runs-degrade-eval-performance-in-reinforce-based-packing-policies.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/pillow0705/semantic/frontier--three-sphere-tangency-generates-physically-valid-candidates-analytically-in-sequential-packing.md (100%)
 rename {skills => .agents/skills}/physics/computational-physics/pillow0705/semantic/frontier--two-stage-collapse-model-predicts-training-failure-in-rl-sphere-packing.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/episodic/adaptation--apply-the-explicit-review-rubric-to-the-manuscript-not-to-the-rubric-itself.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/episodic/adaptation--benchmark-and-frontier-gap-scan-for-review-articles.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/episodic/adaptation--hackathon-interactive-webapp-as-differentiator.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/episodic/adaptation--hysteresis-scanning-curve-validation.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/episodic/anomalous--detecting-structural-model-mismatch-via-seasonal-dynamics.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/constraint-failure--grace-resolution-vs-rock-moisture-scale-mismatch.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/constraint-failure--hydrology-model-integration-of-rock-moisture.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/constraint-failure--llm-agent-fortran-refactoring.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/constraint-failure--ocean-color-vs-land-surface-products-for-coastal-turbidity.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/constraint-failure--reviewer-flag-unnamed-datasets-require-named-citations.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/constraint-failure--sap-flow-upscaling-uncertainty-into-likelihood.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/no-change--diagnosing-overconfidence-induced-research-stall.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/no-change--enforcing-output-over-tool-optimization-in-research-systems.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/no-change--forcing-research-decomposition-under-vague-problem-definition.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/operator-fail--cross-cutting-validation-four-variable-types.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/operator-fail--dual-porosity-convergence-tuning.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/operator-fail--ensemble-kalman-filter-smap-grace-localization.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/operator-fail--factorial-experiments-to-isolate-rock-vs-groundwater-contributions.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/operator-fail--mualem-independent-domain-path-dependent-psi-stem.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/operator-fail--three-tier-falsifiability-success-criteria-rock-moisture.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/tie--distinguishing-parameter-error-vs-structural-error-in-models.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/tie--exploiting-cross-domain-concepts-for-hydrological-innovation.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/tie--gmd-vs-james-vs-ai-venue-for-ai-agents-for-esm.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/tie--lightweight-classifier-over-finetuning-for-foundation-embeddings.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/tie--prioritizing-seasonal-patterns-over-pointwise-fit-in-geoscience.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/tie--sce-ua-vs-mcmc-dream-for-hydraulic-parameter-calibration.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/procedural/tie--sequential-calibration-to-manage-parameter-compounding.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--bayesian-parameter-estimation-is-data-assimilation-naming.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--benchmarking-is-a-core-axis-not-a-side-note-in-modern-lsm-reviews.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--blanket-superiority-claims-for-ml-in-environmental-modelling-need-benchmark-and-mechanism-qualifiers.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--hugo-fabricated-sapwood-50-percent-daily-transpiration.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--lsm-grid-resolution-is-a-design-choice-not-a-physical-limit.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--noah-mp-wood-moisture-initialization.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--sif-light-vs-dark-reactions-decoupling-under-stress.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--split-sample-vs-leave-one-out-cross-validation-distinction.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/correction--temporal-axis-for-lsm-domain-figures-correct-semantic-interpretation.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/frontier--alphaearth-10m-64band-annual-embeddings.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/frontier--deficit-based-mass-balance-lower-bound-sbedrock.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/frontier--dual-anonymous-peer-review-formatting-finesst.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/frontier--mcp-physics-models-novelty-gap.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/frontier--meteor-detection-radon-transform-vs-cnn-pipeline.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/frontier--rock-moisture-storage-capacity-scaling.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/frontier--van-genuchten-hysteresis-stem-water-curve.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/non-public--hunga-tonga-lamb-wave-two-mode-propagation.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/non-public--mengcheng-meteor-radar-double-gaussian-dual-peak.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/non-public--noah-mp-dual-porosity-bedrock-extension-spec.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/non-public--noah-mp-ncar-hpc-allocation-cpu-budget.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/non-public--pelletier-regolith-thickness-dataset-for-noahmp-depth.md (100%)
 rename {skills => .agents/skills}/physics/geophysics/ktwu01/semantic/non-public--phase-1-prior-work-line-counts-fortran.md (100%)
 rename {skills => .agents/skills}/physics/medical-physics/wangmengxiao319/procedural/tie--varp-checkpoint-reuse-to-skip-s1-re-simulation.md (100%)
 rename {skills => .agents/skills}/physics/medical-physics/wangmengxiao319/semantic/correction--cardiac-tissue-limit-cycle-is-pacing-site-specific.md (100%)
 rename {skills => .agents/skills}/physics/medical-physics/wangmengxiao319/semantic/correction--ventricular-varp-pacing-sites-are-endocardial-not-epicardial.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/episodic/adaptation--ssj-conductance-first-visualization-over-raw-spectra.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/episodic/anomalous--fte-satellite-relative-motion-reference-frame.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/episodic/failure--fte-satellite-reference-frame-fte-advection-dominates-spacecraft-motion.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/procedural/constraint-failure--bipolar-bn-as-sampling-artifact-of-moving-structure.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/semantic/correction--bn-0-contour-is-not-the-flux-rope-cross-section.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/semantic/correction--bn-0-is-a-projected-identification-proxy-not-a-true-3d-magnetic-axis.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/semantic/correction--robinson-1987-conductance-formula-full-vs-simplified-form-discrepancy.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/semantic/correction--robinson-formula-yields-height-integrated-conductance-not-local-conductivity.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/semantic/frontier--instrument-provenance-determines-data-selection-in-multi-instrument-reproduction.md (100%)
 rename {skills => .agents/skills}/physics/space-physics/kkeerb/semantic/frontier--two-layer-data-source-model-for-satellite-plotting-tools.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/genomics/ergan-shang/episodic/adaptation--trvae-unpaired-condition-generation.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/genomics/ergan-shang/procedural/constraint-failure--slaf-large-scale-anndata-merging.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/genomics/jamesyu420/procedural/operator-fail--permutation-invariant-program-recovery-protocol.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/genomics/jamesyu420/procedural/tie--graph-guided-spatial-programs-need-uq.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--condition-module-can-corrupt-ranking-evaluation-indirectly.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--order-split-diagnosis-with-test-only-entities.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--distinguish-auroc-auprc-from-ranking-metrics-in-combination-prediction.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--do-not-overclaim-biological-priors-when-features-are-random-or-simplified.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--open-order-generalization-as-order-composition.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--picloss-edge-source-must-follow-split-regime.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--direct-composed-imitation-loss-for-order-experts.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--leave-one-out-contextualization-for-combination-entity-conditioning.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--lexicographic-order-decomposition-for-unseen-k.md (100%)
 rename {skills => .agents/skills}/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--single-positive-versus-multi-positive-mrr-choice.md (100%)
 rename {skills => .agents/skills}/statistics/computation/jamesyu420/procedural/operator-fail--gp-kernel-psd-and-jitter-check.md (100%)
 rename {skills => .agents/skills}/statistics/machine-learning/ergan-shang/procedural/constraint-failure--cross-validation-by-column-masking-in-rasch.md (100%)
 rename {skills => .agents/skills}/statistics/machine-learning/ergan-shang/semantic/correction--bernoulli-logprob-loss-interpretation-in-rasch.md (100%)
 rename {skills => .agents/skills}/statistics/methodology/ergan-shang/procedural/constraint-failure--auc-computation-with-masked-matrix-data.md (100%)
 rename {skills => .agents/skills}/statistics/methodology/jamesyu420/episodic/adaptation--mil-to-sgmm-when-spatial-clustering-matters.md (100%)
 rename {skills => .agents/skills}/statistics/methodology/jamesyu420/semantic/non-public--psmil-bmle-smle-partial-labeling.md (100%)
 rename {skills => .agents/skills}/statistics/methodology/jamesyu420/semantic/non-public--sgmm-spatially-varying-mixture-proportions.md (100%)
 create mode 100644 AGENT.md
 rename readme.md => README.md (86%)

diff --git a/skills/_template-episodic.md b/.agents/skills/_template-episodic.md
similarity index 89%
rename from skills/_template-episodic.md
rename to .agents/skills/_template-episodic.md
index 8c7a269..818fc16 100644
--- a/skills/_template-episodic.md
+++ b/.agents/skills/_template-episodic.md
@@ -3,7 +3,7 @@ name: your-skill-name
 memory_type: episodic
 subtype: failure                               # failure | adaptation | anomalous
 domain: physics                                # physics | mathematics | computer-science | quantitative-biology | statistics | eess | economics | quantitative-finance
-subdomain: geophysics                          # arXiv-aligned subdomain (see skills/<domain>/ for options)
+subdomain: geophysics                          # arXiv-aligned subdomain (see .agents/skills/<domain>/ for options)
 contributor: your-github-username
 ---
 
@@ -19,8 +19,8 @@ contributor: your-github-username
     anomalous  — Expected A, observed B — turned out important
 
   File placement:
-    skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
-    Example: skills/physics/geophysics/jdoe/episodic/failure--gradient-explosion-under-fp16.md
+    .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+    Example: .agents/skills/physics/geophysics/jdoe/episodic/failure--gradient-explosion-under-fp16.md
 
   Instructions:
     1. Fill in all frontmatter fields above.
diff --git a/skills/_template-procedural.md b/.agents/skills/_template-procedural.md
similarity index 89%
rename from skills/_template-procedural.md
rename to .agents/skills/_template-procedural.md
index b98f5e6..c0b58ac 100644
--- a/skills/_template-procedural.md
+++ b/.agents/skills/_template-procedural.md
@@ -3,7 +3,7 @@ name: your-skill-name
 memory_type: procedural
 subtype: tie                                   # tie | no-change | constraint-failure | operator-fail
 domain: physics                                # physics | mathematics | computer-science | quantitative-biology | statistics | eess | economics | quantitative-finance
-subdomain: geophysics                          # arXiv-aligned subdomain (see skills/<domain>/ for options)
+subdomain: geophysics                          # arXiv-aligned subdomain (see .agents/skills/<domain>/ for options)
 contributor: your-github-username
 ---
 
@@ -20,8 +20,8 @@ contributor: your-github-username
     operator-fail    — Correct approach selected but execution fails
 
   File placement:
-    skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
-    Example: skills/physics/geophysics/jdoe/procedural/tie--exploiting-cross-domain-concepts.md
+    .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+    Example: .agents/skills/physics/geophysics/jdoe/procedural/tie--exploiting-cross-domain-concepts.md
 
   Instructions:
     1. Fill in all frontmatter fields above.
diff --git a/skills/_template-semantic.md b/.agents/skills/_template-semantic.md
similarity index 88%
rename from skills/_template-semantic.md
rename to .agents/skills/_template-semantic.md
index 01831c3..6c8466b 100644
--- a/skills/_template-semantic.md
+++ b/.agents/skills/_template-semantic.md
@@ -3,7 +3,7 @@ name: your-skill-name
 memory_type: semantic
 subtype: correction                            # frontier | non-public | correction
 domain: physics                                # physics | mathematics | computer-science | quantitative-biology | statistics | eess | economics | quantitative-finance
-subdomain: geophysics                          # arXiv-aligned subdomain (see skills/<domain>/ for options)
+subdomain: geophysics                          # arXiv-aligned subdomain (see .agents/skills/<domain>/ for options)
 contributor: your-github-username
 ---
 
@@ -19,8 +19,8 @@ contributor: your-github-username
     correction  — LLM actively gets this wrong — confident but incorrect default belief
 
   File placement:
-    skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
-    Example: skills/physics/geophysics/jdoe/semantic/correction--lsm-grid-resolution-is-a-design-choice.md
+    .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+    Example: .agents/skills/physics/geophysics/jdoe/semantic/correction--lsm-grid-resolution-is-a-design-choice.md
 
   Instructions:
     1. Fill in all frontmatter fields above.
diff --git a/skills/_template.md b/.agents/skills/_template.md
similarity index 96%
rename from skills/_template.md
rename to .agents/skills/_template.md
index 3420306..2bc177e 100644
--- a/skills/_template.md
+++ b/.agents/skills/_template.md
@@ -16,7 +16,7 @@ status: draft                                # leave as draft; reviewer will upd
   1. Delete these comment blocks before submitting.
   2. Fill in ALL required frontmatter fields above.
   3. Complete each section below. "Required" sections must be present.
-  4. Run `python utils/tools/validate.py skills/<domain>/<subdomain>/<your-skill>.md` before opening a PR.
+  4. Run `python utils/tools/validate.py .agents/skills/<domain>/<subdomain>/<your-skill>.md` before opening a PR.
   5. See utils/SKILL_SCHEMA.md for full field documentation.
 
   NOTE: The primary contribution method is now skills via /researchskills-extract.
diff --git a/skills/clinical-medicine/.gitkeep b/.agents/skills/clinical-medicine/.gitkeep
similarity index 100%
rename from skills/clinical-medicine/.gitkeep
rename to .agents/skills/clinical-medicine/.gitkeep
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--benchmark-prompt-must-strip-gold-answers.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--benchmark-prompt-must-strip-gold-answers.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--benchmark-prompt-must-strip-gold-answers.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--benchmark-prompt-must-strip-gold-answers.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--noise-first-item-triage-before-cross-round-interpretation.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--noise-first-item-triage-before-cross-round-interpretation.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--noise-first-item-triage-before-cross-round-interpretation.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--noise-first-item-triage-before-cross-round-interpretation.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--provenance-validation-when-incident-source-links-are-unreachable.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--provenance-validation-when-incident-source-links-are-unreachable.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--provenance-validation-when-incident-source-links-are-unreachable.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--provenance-validation-when-incident-source-links-are-unreachable.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-legacy-digital-options-around-talent-led-option-formation.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-legacy-digital-options-around-talent-led-option-formation.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-legacy-digital-options-around-talent-led-option-formation.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-legacy-digital-options-around-talent-led-option-formation.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-novelty-claims-when-prior-literature-already-occupies-the-measurement-contribution.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-novelty-claims-when-prior-literature-already-occupies-the-measurement-contribution.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-novelty-claims-when-prior-literature-already-occupies-the-measurement-contribution.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--reframe-novelty-claims-when-prior-literature-already-occupies-the-measurement-contribution.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--theory-grounded-supplementation-when-open-question-banks-are-inaccessible.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--theory-grounded-supplementation-when-open-question-banks-are-inaccessible.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--theory-grounded-supplementation-when-open-question-banks-are-inaccessible.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/adaptation--theory-grounded-supplementation-when-open-question-banks-are-inaccessible.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--convergent-wrong-distractor-signals-shared-benchmark-structure.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--convergent-wrong-distractor-signals-shared-benchmark-structure.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--convergent-wrong-distractor-signals-shared-benchmark-structure.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--convergent-wrong-distractor-signals-shared-benchmark-structure.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--fixed-effects-can-destroy-identification-when-within-firm-signal-is-thin.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--fixed-effects-can-destroy-identification-when-within-firm-signal-is-thin.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--fixed-effects-can-destroy-identification-when-within-firm-signal-is-thin.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--fixed-effects-can-destroy-identification-when-within-firm-signal-is-thin.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--no-single-k-cluster-validation-for-exploratory-incident-taxonomies.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--no-single-k-cluster-validation-for-exploratory-incident-taxonomies.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--no-single-k-cluster-validation-for-exploratory-incident-taxonomies.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/anomalous--no-single-k-cluster-validation-for-exploratory-incident-taxonomies.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/failure--prefer-mechanism-grounded-metaphors-over-spectacular-but-brittle-proposal-framing.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/failure--prefer-mechanism-grounded-metaphors-over-spectacular-but-brittle-proposal-framing.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/failure--prefer-mechanism-grounded-metaphors-over-spectacular-but-brittle-proposal-framing.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/episodic/failure--prefer-mechanism-grounded-metaphors-over-spectacular-but-brittle-proposal-framing.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ai-incident-theory-requires-an-adversarial-intent-dimension.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ai-incident-theory-requires-an-adversarial-intent-dimension.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ai-incident-theory-requires-an-adversarial-intent-dimension.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ai-incident-theory-requires-an-adversarial-intent-dimension.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ambidexterity-product-is-combined-not-balance.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ambidexterity-product-is-combined-not-balance.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ambidexterity-product-is-combined-not-balance.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--ambidexterity-product-is-combined-not-balance.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--contamination-aware-benchmark-interpretation-for-management-qa.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--contamination-aware-benchmark-interpretation-for-management-qa.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--contamination-aware-benchmark-interpretation-for-management-qa.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--contamination-aware-benchmark-interpretation-for-management-qa.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--digital-option-theory-is-not-finance-binary-options.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--digital-option-theory-is-not-finance-binary-options.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--digital-option-theory-is-not-finance-binary-options.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--digital-option-theory-is-not-finance-binary-options.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--do-not-collapse-governance-digitalization-into-management-informatization.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--do-not-collapse-governance-digitalization-into-management-informatization.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--do-not-collapse-governance-digitalization-into-management-informatization.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--do-not-collapse-governance-digitalization-into-management-informatization.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--non-significant-endogeneity-tests-do-not-prove-exogeneity.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--non-significant-endogeneity-tests-do-not-prove-exogeneity.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--non-significant-endogeneity-tests-do-not-prove-exogeneity.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--non-significant-endogeneity-tests-do-not-prove-exogeneity.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--nonlinear-endogeneity-prefers-control-function-over-naive-2sls.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--nonlinear-endogeneity-prefers-control-function-over-naive-2sls.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--nonlinear-endogeneity-prefers-control-function-over-naive-2sls.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--nonlinear-endogeneity-prefers-control-function-over-naive-2sls.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--novelty-in-agentic-is-should-be-framed-as-calibrated-extension-not-paradigm-pioneer.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--novelty-in-agentic-is-should-be-framed-as-calibrated-extension-not-paradigm-pioneer.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--novelty-in-agentic-is-should-be-framed-as-calibrated-extension-not-paradigm-pioneer.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--novelty-in-agentic-is-should-be-framed-as-calibrated-extension-not-paradigm-pioneer.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--rebalance-benchmark-datasets-by-subdimension-not-by-total-item-count.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--rebalance-benchmark-datasets-by-subdimension-not-by-total-item-count.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--rebalance-benchmark-datasets-by-subdimension-not-by-total-item-count.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/correction--rebalance-benchmark-datasets-by-subdimension-not-by-total-item-count.md
diff --git a/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/frontier--tokenizer-level-name-failures-can-masquerade-as-knowledge-gaps.md b/.agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/frontier--tokenizer-level-name-failures-can-masquerade-as-knowledge-gaps.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/frontier--tokenizer-level-name-failures-can-masquerade-as-knowledge-gaps.md
rename to .agents/skills/computer-science/artificial-intelligence/clarence-zhewang/semantic/frontier--tokenizer-level-name-failures-can-masquerade-as-knowledge-gaps.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/episodic/adaptation--freeing-memory-by-dropping-non-backprop-models.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/episodic/adaptation--freeing-memory-by-dropping-non-backprop-models.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/episodic/adaptation--freeing-memory-by-dropping-non-backprop-models.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/episodic/adaptation--freeing-memory-by-dropping-non-backprop-models.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/procedural/constraint-failure--token-level-grpo-mask-invariance.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/constraint-failure--token-level-grpo-mask-invariance.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/procedural/constraint-failure--token-level-grpo-mask-invariance.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/constraint-failure--token-level-grpo-mask-invariance.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--cuda-memory-gap-fragmentation-diagnosis.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--cuda-memory-gap-fragmentation-diagnosis.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--cuda-memory-gap-fragmentation-diagnosis.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--cuda-memory-gap-fragmentation-diagnosis.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--detaching-external-model-outputs-before-training.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--detaching-external-model-outputs-before-training.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--detaching-external-model-outputs-before-training.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--detaching-external-model-outputs-before-training.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--grpo-numerical-stability-kl-divergence.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--grpo-numerical-stability-kl-divergence.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--grpo-numerical-stability-kl-divergence.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/operator-fail--grpo-numerical-stability-kl-divergence.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--bnpo-vs-dapo-loss-scaling-choice.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--bnpo-vs-dapo-loss-scaling-choice.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--bnpo-vs-dapo-loss-scaling-choice.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--bnpo-vs-dapo-loss-scaling-choice.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--gradient-checkpointing-target-selection.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--gradient-checkpointing-target-selection.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--gradient-checkpointing-target-selection.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--gradient-checkpointing-target-selection.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--llm-benchmarking-regression-heuristics.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--llm-benchmarking-regression-heuristics.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--llm-benchmarking-regression-heuristics.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/procedural/tie--llm-benchmarking-regression-heuristics.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--reproducibility-seed-limitations-in-pytorch.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--reproducibility-seed-limitations-in-pytorch.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--reproducibility-seed-limitations-in-pytorch.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--reproducibility-seed-limitations-in-pytorch.md
diff --git a/skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--tokenizer-invariance-during-model-updates.md b/.agents/skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--tokenizer-invariance-during-model-updates.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--tokenizer-invariance-during-model-updates.md
rename to .agents/skills/computer-science/artificial-intelligence/ergan-shang/semantic/correction--tokenizer-invariance-during-model-updates.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--jagged-frontier-verifiability-gates-automation.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--jagged-frontier-verifiability-gates-automation.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--jagged-frontier-verifiability-gates-automation.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--jagged-frontier-verifiability-gates-automation.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--kantian-categories-wrong-granularity-for-token-level-priors.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--kantian-categories-wrong-granularity-for-token-level-priors.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--kantian-categories-wrong-granularity-for-token-level-priors.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--kantian-categories-wrong-granularity-for-token-level-priors.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--post-cutoff-model-partial-recognition-trap.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--post-cutoff-model-partial-recognition-trap.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--post-cutoff-model-partial-recognition-trap.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/constraint-failure--post-cutoff-model-partial-recognition-trap.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--direction-negation-as-skill-signal-in-conversation-logs.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--direction-negation-as-skill-signal-in-conversation-logs.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--direction-negation-as-skill-signal-in-conversation-logs.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--direction-negation-as-skill-signal-in-conversation-logs.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--embedding-structure-layer-question-before-injecting.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--embedding-structure-layer-question-before-injecting.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--embedding-structure-layer-question-before-injecting.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--embedding-structure-layer-question-before-injecting.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--evaluate-research-skills-without-benchmark.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--evaluate-research-skills-without-benchmark.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--evaluate-research-skills-without-benchmark.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--evaluate-research-skills-without-benchmark.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--paper-trajectory-is-higher-signal-than-final-paper.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--paper-trajectory-is-higher-signal-than-final-paper.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--paper-trajectory-is-higher-signal-than-final-paper.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/no-change--paper-trajectory-is-higher-signal-than-final-paper.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--conference-topic-from-x-to-y-framing.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--conference-topic-from-x-to-y-framing.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--conference-topic-from-x-to-y-framing.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--conference-topic-from-x-to-y-framing.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--kantian-category-implementation-start-narrow.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--kantian-category-implementation-start-narrow.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--kantian-category-implementation-start-narrow.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--kantian-category-implementation-start-narrow.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--research-skill-format-four-section-template.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--research-skill-format-four-section-template.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--research-skill-format-four-section-template.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/operator-fail--research-skill-format-four-section-template.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--direction-not-capacity-is-binding-in-ai-research-2026.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--direction-not-capacity-is-binding-in-ai-research-2026.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--direction-not-capacity-is-binding-in-ai-research-2026.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--direction-not-capacity-is-binding-in-ai-research-2026.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--mechanistic-interpretability-is-enabling-condition-not-standalone-direction.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--mechanistic-interpretability-is-enabling-condition-not-standalone-direction.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--mechanistic-interpretability-is-enabling-condition-not-standalone-direction.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--mechanistic-interpretability-is-enabling-condition-not-standalone-direction.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-direction-closed-loop-ordering.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-direction-closed-loop-ordering.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-direction-closed-loop-ordering.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-direction-closed-loop-ordering.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-skill-definition-three-conditions.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-skill-definition-three-conditions.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-skill-definition-three-conditions.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--research-skill-definition-three-conditions.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--shallow-validated-skill-library-beats-deep-autogenerated.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--shallow-validated-skill-library-beats-deep-autogenerated.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--shallow-validated-skill-library-beats-deep-autogenerated.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/procedural/tie--shallow-validated-skill-library-beats-deep-autogenerated.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-domain-coverage-inverse-proportional-to-skill-value.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-domain-coverage-inverse-proportional-to-skill-value.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-domain-coverage-inverse-proportional-to-skill-value.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-domain-coverage-inverse-proportional-to-skill-value.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-negative-transfer-mechanism.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-negative-transfer-mechanism.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-negative-transfer-mechanism.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/correction--skillsbench-negative-transfer-mechanism.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--bittokens-ieee754-single-token-arithmetic.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--bittokens-ieee754-single-token-arithmetic.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--bittokens-ieee754-single-token-arithmetic.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--bittokens-ieee754-single-token-arithmetic.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--deepseek-engram-separating-computation-from-retrieval.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--deepseek-engram-separating-computation-from-retrieval.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--deepseek-engram-separating-computation-from-retrieval.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--deepseek-engram-separating-computation-from-retrieval.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--nemotron-h-hybrid-92-percent-mamba-replacement.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--nemotron-h-hybrid-92-percent-mamba-replacement.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--nemotron-h-hybrid-92-percent-mamba-replacement.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--nemotron-h-hybrid-92-percent-mamba-replacement.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--sakana-ai-scientist-nature-publication-march-2026.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--sakana-ai-scientist-nature-publication-march-2026.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--sakana-ai-scientist-nature-publication-march-2026.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--sakana-ai-scientist-nature-publication-march-2026.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-creator-quality-dominates-executor-quality.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-creator-quality-dominates-executor-quality.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-creator-quality-dominates-executor-quality.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-creator-quality-dominates-executor-quality.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-hierarchical-skill-composition-fails.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-hierarchical-skill-composition-fails.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-hierarchical-skill-composition-fails.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-hierarchical-skill-composition-fails.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-skill-judgment-matters-more-than-skill-creation.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-skill-judgment-matters-more-than-skill-creation.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-skill-judgment-matters-more-than-skill-creation.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillcraft-skill-judgment-matters-more-than-skill-creation.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillsbench-human-vs-self-generated-skill-gap.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillsbench-human-vs-self-generated-skill-gap.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillsbench-human-vs-self-generated-skill-gap.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--skillsbench-human-vs-self-generated-skill-gap.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--tokenization-right-to-left-for-arithmetic.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--tokenization-right-to-left-for-arithmetic.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--tokenization-right-to-left-for-arithmetic.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--tokenization-right-to-left-for-arithmetic.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--ttrl-test-time-rl-without-labels.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--ttrl-test-time-rl-without-labels.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--ttrl-test-time-rl-without-labels.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--ttrl-test-time-rl-without-labels.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--xval-continuous-number-embedding-scaling.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--xval-continuous-number-embedding-scaling.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--xval-continuous-number-embedding-scaling.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/frontier--xval-continuous-number-embedding-scaling.md
diff --git a/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/non-public--conversation-noise-floor-3-to-5-signal-turns-per-200.md b/.agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/non-public--conversation-noise-floor-3-to-5-signal-turns-per-200.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/hhhhhejia/semantic/non-public--conversation-noise-floor-3-to-5-signal-turns-per-200.md
rename to .agents/skills/computer-science/artificial-intelligence/hhhhhejia/semantic/non-public--conversation-noise-floor-3-to-5-signal-turns-per-200.md
diff --git a/skills/computer-science/artificial-intelligence/ktwu01/procedural/tie--ai-agent-discipline-operational-failure-modes-to-guard-against.md b/.agents/skills/computer-science/artificial-intelligence/ktwu01/procedural/tie--ai-agent-discipline-operational-failure-modes-to-guard-against.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/ktwu01/procedural/tie--ai-agent-discipline-operational-failure-modes-to-guard-against.md
rename to .agents/skills/computer-science/artificial-intelligence/ktwu01/procedural/tie--ai-agent-discipline-operational-failure-modes-to-guard-against.md
diff --git a/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--base-model-empty-responses-from-chat-format-prompts.md b/.agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--base-model-empty-responses-from-chat-format-prompts.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--base-model-empty-responses-from-chat-format-prompts.md
rename to .agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--base-model-empty-responses-from-chat-format-prompts.md
diff --git a/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--comparison-table-averaging-formula-audit.md b/.agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--comparison-table-averaging-formula-audit.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--comparison-table-averaging-formula-audit.md
rename to .agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--comparison-table-averaging-formula-audit.md
diff --git a/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--correct-answer-rarely-has-flawed-process-in-rl-models.md b/.agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--correct-answer-rarely-has-flawed-process-in-rl-models.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--correct-answer-rarely-has-flawed-process-in-rl-models.md
rename to .agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--correct-answer-rarely-has-flawed-process-in-rl-models.md
diff --git a/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--rank-ordering-guarantee-dual-advantage-rl.md b/.agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--rank-ordering-guarantee-dual-advantage-rl.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--rank-ordering-guarantee-dual-advantage-rl.md
rename to .agents/skills/computer-science/artificial-intelligence/tanzelin430/episodic/anomalous--rank-ordering-guarantee-dual-advantage-rl.md
diff --git a/skills/computer-science/artificial-intelligence/tanzelin430/semantic/non-public--verification-habit-emerges-from-process-supervision.md b/.agents/skills/computer-science/artificial-intelligence/tanzelin430/semantic/non-public--verification-habit-emerges-from-process-supervision.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/tanzelin430/semantic/non-public--verification-habit-emerges-from-process-supervision.md
rename to .agents/skills/computer-science/artificial-intelligence/tanzelin430/semantic/non-public--verification-habit-emerges-from-process-supervision.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--mcculloch-pitts-is-best-framed-as-the-brain-written-as-formula.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--mcculloch-pitts-is-best-framed-as-the-brain-written-as-formula.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--mcculloch-pitts-is-best-framed-as-the-brain-written-as-formula.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--mcculloch-pitts-is-best-framed-as-the-brain-written-as-formula.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--rebuttal-efficiency-comes-from-defining-the-metric-boundary.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--rebuttal-efficiency-comes-from-defining-the-metric-boundary.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--rebuttal-efficiency-comes-from-defining-the-metric-boundary.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/episodic/adaptation--rebuttal-efficiency-comes-from-defining-the-metric-boundary.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--arbitrary-size-gnn-claims-need-scaling-and-input-contract-tests.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--arbitrary-size-gnn-claims-need-scaling-and-input-contract-tests.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--arbitrary-size-gnn-claims-need-scaling-and-input-contract-tests.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--arbitrary-size-gnn-claims-need-scaling-and-input-contract-tests.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--inference-time-must-define-initial-solve-versus-post-optimization.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--inference-time-must-define-initial-solve-versus-post-optimization.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--inference-time-must-define-initial-solve-versus-post-optimization.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--inference-time-must-define-initial-solve-versus-post-optimization.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--internal-benchmark-results-should-not-be-overfit-into-public-scientific-claims.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--internal-benchmark-results-should-not-be-overfit-into-public-scientific-claims.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--internal-benchmark-results-should-not-be-overfit-into-public-scientific-claims.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--internal-benchmark-results-should-not-be-overfit-into-public-scientific-claims.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--llm-agent-evaluation-needs-stability-not-just-best-run.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--llm-agent-evaluation-needs-stability-not-just-best-run.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--llm-agent-evaluation-needs-stability-not-just-best-run.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--llm-agent-evaluation-needs-stability-not-just-best-run.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--placement-benchmark-splits-must-avoid-design-family-leakage.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--placement-benchmark-splits-must-avoid-design-family-leakage.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--placement-benchmark-splits-must-avoid-design-family-leakage.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--placement-benchmark-splits-must-avoid-design-family-leakage.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--same-parameter-ablation-for-model-compression-claims.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--same-parameter-ablation-for-model-compression-claims.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--same-parameter-ablation-for-model-compression-claims.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--same-parameter-ablation-for-model-compression-claims.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--semi-supervised-claims-require-label-budget-accounting.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--semi-supervised-claims-require-label-budget-accounting.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--semi-supervised-claims-require-label-budget-accounting.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--semi-supervised-claims-require-label-budget-accounting.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--sequence-level-agent-training-needs-outcome-credit-not-only-token-kl.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--sequence-level-agent-training-needs-outcome-credit-not-only-token-kl.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--sequence-level-agent-training-needs-outcome-credit-not-only-token-kl.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--sequence-level-agent-training-needs-outcome-credit-not-only-token-kl.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--thinking-budget-is-a-hidden-confounder-in-agent-benchmarks.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--thinking-budget-is-a-hidden-confounder-in-agent-benchmarks.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--thinking-budget-is-a-hidden-confounder-in-agent-benchmarks.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--thinking-budget-is-a-hidden-confounder-in-agent-benchmarks.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-driven-placement-must-end-in-sta-metrics.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-driven-placement-must-end-in-sta-metrics.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-driven-placement-must-end-in-sta-metrics.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-driven-placement-must-end-in-sta-metrics.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-surrogates-need-path-topology-not-just-geometry.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-surrogates-need-path-topology-not-just-geometry.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-surrogates-need-path-topology-not-just-geometry.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/constraint-failure--timing-surrogates-need-path-topology-not-just-geometry.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--dead-end-experiments-should-be-kept-as-reviewer-risk-memory.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--dead-end-experiments-should-be-kept-as-reviewer-risk-memory.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--dead-end-experiments-should-be-kept-as-reviewer-risk-memory.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--dead-end-experiments-should-be-kept-as-reviewer-risk-memory.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--reviewer-concerns-should-be-clustered-into-experiment-design-fixes.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--reviewer-concerns-should-be-clustered-into-experiment-design-fixes.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--reviewer-concerns-should-be-clustered-into-experiment-design-fixes.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/no-change--reviewer-concerns-should-be-clustered-into-experiment-design-fixes.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-chapters-need-a-single-philosophical-thread.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-chapters-need-a-single-philosophical-thread.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-chapters-need-a-single-philosophical-thread.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-chapters-need-a-single-philosophical-thread.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-popularization-should-trim-detours-not-dilute-concepts.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-popularization-should-trim-detours-not-dilute-concepts.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-popularization-should-trim-detours-not-dilute-concepts.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--ai-history-popularization-should-trim-detours-not-dilute-concepts.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--differentiable-ilp-gradients-require-bias-and-variance-audits.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--differentiable-ilp-gradients-require-bias-and-variance-audits.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--differentiable-ilp-gradients-require-bias-and-variance-audits.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--differentiable-ilp-gradients-require-bias-and-variance-audits.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--kv-eviction-should-be-evaluated-by-output-drift-not-only-retained-token-score.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--kv-eviction-should-be-evaluated-by-output-drift-not-only-retained-token-score.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--kv-eviction-should-be-evaluated-by-output-drift-not-only-retained-token-score.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--kv-eviction-should-be-evaluated-by-output-drift-not-only-retained-token-score.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--neural-milp-solutions-need-solver-interface-specification.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--neural-milp-solutions-need-solver-interface-specification.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--neural-milp-solutions-need-solver-interface-specification.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--neural-milp-solutions-need-solver-interface-specification.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--policy-gradient-explanations-should-start-from-trajectory-probability.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--policy-gradient-explanations-should-start-from-trajectory-probability.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--policy-gradient-explanations-should-start-from-trajectory-probability.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--policy-gradient-explanations-should-start-from-trajectory-probability.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--related-work-baselines-are-required-when-contrasting-learning-paradigms.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--related-work-baselines-are-required-when-contrasting-learning-paradigms.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--related-work-baselines-are-required-when-contrasting-learning-paradigms.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--related-work-baselines-are-required-when-contrasting-learning-paradigms.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--timing-driven-global-placement-needs-gradient-usefulness-check.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--timing-driven-global-placement-needs-gradient-usefulness-check.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--timing-driven-global-placement-needs-gradient-usefulness-check.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--timing-driven-global-placement-needs-gradient-usefulness-check.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--training-curve-evidence-for-learning-regime-comparisons.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--training-curve-evidence-for-learning-regime-comparisons.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--training-curve-evidence-for-learning-regime-comparisons.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/operator-fail--training-curve-evidence-for-learning-regime-comparisons.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-eda-paper-claims-need-proxy-to-signoff-causal-chain.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-eda-paper-claims-need-proxy-to-signoff-causal-chain.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-eda-paper-claims-need-proxy-to-signoff-causal-chain.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-eda-paper-claims-need-proxy-to-signoff-causal-chain.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-history-book-positioning-should-be-history-of-ideas-not-event-chronicle.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-history-book-positioning-should-be-history-of-ideas-not-event-chronicle.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-history-book-positioning-should-be-history-of-ideas-not-event-chronicle.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--ai-history-book-positioning-should-be-history-of-ideas-not-event-chronicle.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--benchmark-release-from-internal-diagnostic-should-be-positioned-conservatively.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--benchmark-release-from-internal-diagnostic-should-be-positioned-conservatively.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--benchmark-release-from-internal-diagnostic-should-be-positioned-conservatively.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--benchmark-release-from-internal-diagnostic-should-be-positioned-conservatively.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--entropy-is-a-routing-signal-for-speculative-reasoning-not-a-theme-word.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--entropy-is-a-routing-signal-for-speculative-reasoning-not-a-theme-word.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--entropy-is-a-routing-signal-for-speculative-reasoning-not-a-theme-word.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--entropy-is-a-routing-signal-for-speculative-reasoning-not-a-theme-word.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--grpo-should-be-framed-as-relative-advantage-without-a-critic.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--grpo-should-be-framed-as-relative-advantage-without-a-critic.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--grpo-should-be-framed-as-relative-advantage-without-a-critic.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--grpo-should-be-framed-as-relative-advantage-without-a-critic.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--hilbert-godel-turing-arc-should-be-used-as-computability-origin-not-name-dropping.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--hilbert-godel-turing-arc-should-be-used-as-computability-origin-not-name-dropping.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--hilbert-godel-turing-arc-should-be-used-as-computability-origin-not-name-dropping.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--hilbert-godel-turing-arc-should-be-used-as-computability-origin-not-name-dropping.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--information-control-systems-should-be-presented-as-three-lenses-not-one-slogan.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--information-control-systems-should-be-presented-as-three-lenses-not-one-slogan.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--information-control-systems-should-be-presented-as-three-lenses-not-one-slogan.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--information-control-systems-should-be-presented-as-three-lenses-not-one-slogan.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--on-policy-distillation-differs-from-ppo-by-supervision-source-and-objective.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--on-policy-distillation-differs-from-ppo-by-supervision-source-and-objective.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--on-policy-distillation-differs-from-ppo-by-supervision-source-and-objective.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--on-policy-distillation-differs-from-ppo-by-supervision-source-and-objective.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--pre-dartmouth-ai-origin-arc-should-be-mechanism-first.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--pre-dartmouth-ai-origin-arc-should-be-mechanism-first.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--pre-dartmouth-ai-origin-arc-should-be-mechanism-first.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--pre-dartmouth-ai-origin-arc-should-be-mechanism-first.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--scholarly-ai-history-needs-modern-llm-return-path.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--scholarly-ai-history-needs-modern-llm-return-path.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--scholarly-ai-history-needs-modern-llm-return-path.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--scholarly-ai-history-needs-modern-llm-return-path.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--task-success-criteria-are-more-important-than-task-count-in-agent-benchmarks.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--task-success-criteria-are-more-important-than-task-count-in-agent-benchmarks.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--task-success-criteria-are-more-important-than-task-count-in-agent-benchmarks.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/procedural/tie--task-success-criteria-are-more-important-than-task-count-in-agent-benchmarks.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--attention-weight-importance-is-not-enough-for-kv-eviction.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--attention-weight-importance-is-not-enough-for-kv-eviction.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--attention-weight-importance-is-not-enough-for-kv-eviction.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--attention-weight-importance-is-not-enough-for-kv-eviction.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--diffusion-for-milp-should-generate-solution-batches-not-marginal-averages.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--diffusion-for-milp-should-generate-solution-batches-not-marginal-averages.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--diffusion-for-milp-should-generate-solution-batches-not-marginal-averages.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--diffusion-for-milp-should-generate-solution-batches-not-marginal-averages.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--llm-judge-binary-scores-do-not-imply-a-hidden-threshold.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--llm-judge-binary-scores-do-not-imply-a-hidden-threshold.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--llm-judge-binary-scores-do-not-imply-a-hidden-threshold.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--llm-judge-binary-scores-do-not-imply-a-hidden-threshold.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--multi-model-speculative-reasoning-is-not-generic-model-collaboration.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--multi-model-speculative-reasoning-is-not-generic-model-collaboration.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--multi-model-speculative-reasoning-is-not-generic-model-collaboration.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--multi-model-speculative-reasoning-is-not-generic-model-collaboration.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--omitted-thinking-parameter-is-not-equivalent-to-high-or-off.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--omitted-thinking-parameter-is-not-equivalent-to-high-or-off.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--omitted-thinking-parameter-is-not-equivalent-to-high-or-off.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--omitted-thinking-parameter-is-not-equivalent-to-high-or-off.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--proof-benchmarks-can-use-olympiad-style-continuous-scoring.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--proof-benchmarks-can-use-olympiad-style-continuous-scoring.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--proof-benchmarks-can-use-olympiad-style-continuous-scoring.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/correction--proof-benchmarks-can-use-olympiad-style-continuous-scoring.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--pass-three-measures-reliability-not-single-run-competence.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--pass-three-measures-reliability-not-single-run-competence.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--pass-three-measures-reliability-not-single-run-competence.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--pass-three-measures-reliability-not-single-run-competence.md
diff --git a/skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--value-space-projection-for-kv-cache-eviction.md b/.agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--value-space-projection-for-kv-cache-eviction.md
similarity index 100%
rename from skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--value-space-projection-for-kv-cache-eviction.md
rename to .agents/skills/computer-science/artificial-intelligence/zijiegeng/semantic/frontier--value-space-projection-for-kv-cache-eviction.md
diff --git a/skills/computer-science/computational-engineering-finance-and-science/ktwu01/episodic/anomalous--incomplete-archived-simulations-can-masquerade-as-scientific-disagreement-in-cross-version-validation.md b/.agents/skills/computer-science/computational-engineering-finance-and-science/ktwu01/episodic/anomalous--incomplete-archived-simulations-can-masquerade-as-scientific-disagreement-in-cross-version-validation.md
similarity index 100%
rename from skills/computer-science/computational-engineering-finance-and-science/ktwu01/episodic/anomalous--incomplete-archived-simulations-can-masquerade-as-scientific-disagreement-in-cross-version-validation.md
rename to .agents/skills/computer-science/computational-engineering-finance-and-science/ktwu01/episodic/anomalous--incomplete-archived-simulations-can-masquerade-as-scientific-disagreement-in-cross-version-validation.md
diff --git a/skills/computer-science/computational-engineering-finance-and-science/ktwu01/semantic/correction--matching-hydrology-options-can-preserve-a-port-while-still-failing-full-system-numerical-equivalence.md b/.agents/skills/computer-science/computational-engineering-finance-and-science/ktwu01/semantic/correction--matching-hydrology-options-can-preserve-a-port-while-still-failing-full-system-numerical-equivalence.md
similarity index 100%
rename from skills/computer-science/computational-engineering-finance-and-science/ktwu01/semantic/correction--matching-hydrology-options-can-preserve-a-port-while-still-failing-full-system-numerical-equivalence.md
rename to .agents/skills/computer-science/computational-engineering-finance-and-science/ktwu01/semantic/correction--matching-hydrology-options-can-preserve-a-port-while-still-failing-full-system-numerical-equivalence.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/episodic/adaptation--invalid-survey-response-removal-before-formal-analysis.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/episodic/adaptation--invalid-survey-response-removal-before-formal-analysis.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/episodic/adaptation--invalid-survey-response-removal-before-formal-analysis.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/episodic/adaptation--invalid-survey-response-removal-before-formal-analysis.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--avoid-causal-or-intervention-claims-in-stress-level-thesis.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--avoid-causal-or-intervention-claims-in-stress-level-thesis.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--avoid-causal-or-intervention-claims-in-stress-level-thesis.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--avoid-causal-or-intervention-claims-in-stress-level-thesis.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--dissertation-hypothesis-alignment-after-changing-h3-to-joint-prediction.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--dissertation-hypothesis-alignment-after-changing-h3-to-joint-prediction.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--dissertation-hypothesis-alignment-after-changing-h3-to-joint-prediction.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--dissertation-hypothesis-alignment-after-changing-h3-to-joint-prediction.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--sampling-wording-must-match-actual-design.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--sampling-wording-must-match-actual-design.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--sampling-wording-must-match-actual-design.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/constraint-failure--sampling-wording-must-match-actual-design.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--preserve-existing-method-structure-when-fixing-math.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--preserve-existing-method-structure-when-fixing-math.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--preserve-existing-method-structure-when-fixing-math.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--preserve-existing-method-structure-when-fixing-math.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--reviewer-summary-should-separate-positive-and-negative-methodological-claims.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--reviewer-summary-should-separate-positive-and-negative-methodological-claims.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--reviewer-summary-should-separate-positive-and-negative-methodological-claims.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/hzzzzzhappy/procedural/operator-fail--reviewer-summary-should-separate-positive-and-negative-methodological-claims.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reduce-multi-object-prompts-to-the-single-swapped-object.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reduce-multi-object-prompts-to-the-single-swapped-object.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reduce-multi-object-prompts-to-the-single-swapped-object.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reduce-multi-object-prompts-to-the-single-swapped-object.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reframe-failed-ranking-claims-as-evidence-of-semantic-recovery.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reframe-failed-ranking-claims-as-evidence-of-semantic-recovery.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reframe-failed-ranking-claims-as-evidence-of-semantic-recovery.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--reframe-failed-ranking-claims-as-evidence-of-semantic-recovery.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--soft-hallucination-example-mining-from-near-miss-object-pairs.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--soft-hallucination-example-mining-from-near-miss-object-pairs.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--soft-hallucination-example-mining-from-near-miss-object-pairs.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/adaptation--soft-hallucination-example-mining-from-near-miss-object-pairs.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/failure--reviewer-pushback-can-reveal-which-criticisms-are-scientific-vs-merely-rhetorical.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/failure--reviewer-pushback-can-reveal-which-criticisms-are-scientific-vs-merely-rhetorical.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/failure--reviewer-pushback-can-reveal-which-criticisms-are-scientific-vs-merely-rhetorical.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/episodic/failure--reviewer-pushback-can-reveal-which-criticisms-are-scientific-vs-merely-rhetorical.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--balanced-hard-baseline-comes-from-the-hypothetical-model-output-not-observed-responses.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--balanced-hard-baseline-comes-from-the-hypothetical-model-output-not-observed-responses.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--balanced-hard-baseline-comes-from-the-hypothetical-model-output-not-observed-responses.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--balanced-hard-baseline-comes-from-the-hypothetical-model-output-not-observed-responses.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--human-score-is-a-separate-target-not-human-accuracy.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--human-score-is-a-separate-target-not-human-accuracy.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--human-score-is-a-separate-target-not-human-accuracy.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--human-score-is-a-separate-target-not-human-accuracy.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--object-level-similarity-not-label-level-similarity.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--object-level-similarity-not-label-level-similarity.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--object-level-similarity-not-label-level-similarity.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/correction--object-level-similarity-not-label-level-similarity.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--alignment-tuning-softens-hallucination-severity-more-reliably-than-rl-style-alignment-in-this-evaluation.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--alignment-tuning-softens-hallucination-severity-more-reliably-than-rl-style-alignment-in-this-evaluation.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--alignment-tuning-softens-hallucination-severity-more-reliably-than-rl-style-alignment-in-this-evaluation.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--alignment-tuning-softens-hallucination-severity-more-reliably-than-rl-style-alignment-in-this-evaluation.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--multi-view-human-alignment-can-increase-variance-without-moving-the-mean-much.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--multi-view-human-alignment-can-increase-variance-without-moving-the-mean-much.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--multi-view-human-alignment-can-increase-variance-without-moving-the-mean-much.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--multi-view-human-alignment-can-increase-variance-without-moving-the-mean-much.md
diff --git a/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--stronger-models-show-fewer-severe-hallucinations-even-when-severe-errors-still-dominate.md b/.agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--stronger-models-show-fewer-severe-hallucinations-even-when-severe-errors-still-dominate.md
similarity index 100%
rename from skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--stronger-models-show-fewer-severe-hallucinations-even-when-severe-errors-still-dominate.md
rename to .agents/skills/computer-science/computer-vision-and-pattern-recognition/xisen-w/semantic/non-public--stronger-models-show-fewer-severe-hallucinations-even-when-severe-errors-still-dominate.md
diff --git a/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--evidence-grounding-in-experiment-reports.md b/.agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--evidence-grounding-in-experiment-reports.md
similarity index 100%
rename from skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--evidence-grounding-in-experiment-reports.md
rename to .agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--evidence-grounding-in-experiment-reports.md
diff --git a/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--research-code-release-blocker-audit.md b/.agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--research-code-release-blocker-audit.md
similarity index 100%
rename from skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--research-code-release-blocker-audit.md
rename to .agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--research-code-release-blocker-audit.md
diff --git a/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--submission-readiness-checks-for-anonymous-papers.md b/.agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--submission-readiness-checks-for-anonymous-papers.md
similarity index 100%
rename from skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--submission-readiness-checks-for-anonymous-papers.md
rename to .agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/constraint-failure--submission-readiness-checks-for-anonymous-papers.md
diff --git a/skills/computer-science/machine-learning/a-green-hand-jack/procedural/no-change--reproducible-job-scripts-for-ml-experiments.md b/.agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/no-change--reproducible-job-scripts-for-ml-experiments.md
similarity index 100%
rename from skills/computer-science/machine-learning/a-green-hand-jack/procedural/no-change--reproducible-job-scripts-for-ml-experiments.md
rename to .agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/no-change--reproducible-job-scripts-for-ml-experiments.md
diff --git a/skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--evidence-backed-work-timeline-blocks.md b/.agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--evidence-backed-work-timeline-blocks.md
similarity index 100%
rename from skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--evidence-backed-work-timeline-blocks.md
rename to .agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--evidence-backed-work-timeline-blocks.md
diff --git a/skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--reverse-chronological-experiment-result-logging.md b/.agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--reverse-chronological-experiment-result-logging.md
similarity index 100%
rename from skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--reverse-chronological-experiment-result-logging.md
rename to .agents/skills/computer-science/machine-learning/a-green-hand-jack/procedural/tie--reverse-chronological-experiment-result-logging.md
diff --git a/skills/computer-science/machine-learning/a-green-hand-jack/semantic/correction--venue-preamble-isolation-for-paper-templates.md b/.agents/skills/computer-science/machine-learning/a-green-hand-jack/semantic/correction--venue-preamble-isolation-for-paper-templates.md
similarity index 100%
rename from skills/computer-science/machine-learning/a-green-hand-jack/semantic/correction--venue-preamble-isolation-for-paper-templates.md
rename to .agents/skills/computer-science/machine-learning/a-green-hand-jack/semantic/correction--venue-preamble-isolation-for-paper-templates.md
diff --git a/skills/computer-science/machine-learning/ktwu01/episodic/adaptation--commit-log-probes-for-retrospective-paper-evaluation.md b/.agents/skills/computer-science/machine-learning/ktwu01/episodic/adaptation--commit-log-probes-for-retrospective-paper-evaluation.md
similarity index 100%
rename from skills/computer-science/machine-learning/ktwu01/episodic/adaptation--commit-log-probes-for-retrospective-paper-evaluation.md
rename to .agents/skills/computer-science/machine-learning/ktwu01/episodic/adaptation--commit-log-probes-for-retrospective-paper-evaluation.md
diff --git a/skills/computer-science/machine-learning/ktwu01/episodic/failure--auto-review-narratives-can-overstate-methodological-rigor-in-paper-diff-assessment.md b/.agents/skills/computer-science/machine-learning/ktwu01/episodic/failure--auto-review-narratives-can-overstate-methodological-rigor-in-paper-diff-assessment.md
similarity index 100%
rename from skills/computer-science/machine-learning/ktwu01/episodic/failure--auto-review-narratives-can-overstate-methodological-rigor-in-paper-diff-assessment.md
rename to .agents/skills/computer-science/machine-learning/ktwu01/episodic/failure--auto-review-narratives-can-overstate-methodological-rigor-in-paper-diff-assessment.md
diff --git a/skills/computer-science/machine-learning/ktwu01/episodic/failure--benchmark-paper-review-found-hidden-data-manuscript-drift.md b/.agents/skills/computer-science/machine-learning/ktwu01/episodic/failure--benchmark-paper-review-found-hidden-data-manuscript-drift.md
similarity index 100%
rename from skills/computer-science/machine-learning/ktwu01/episodic/failure--benchmark-paper-review-found-hidden-data-manuscript-drift.md
rename to .agents/skills/computer-science/machine-learning/ktwu01/episodic/failure--benchmark-paper-review-found-hidden-data-manuscript-drift.md
diff --git a/skills/computer-science/machine-learning/ktwu01/procedural/tie--research-integrity-antipatterns-deceptive-practices-to-reject-in-paper-writing.md b/.agents/skills/computer-science/machine-learning/ktwu01/procedural/tie--research-integrity-antipatterns-deceptive-practices-to-reject-in-paper-writing.md
similarity index 100%
rename from skills/computer-science/machine-learning/ktwu01/procedural/tie--research-integrity-antipatterns-deceptive-practices-to-reject-in-paper-writing.md
rename to .agents/skills/computer-science/machine-learning/ktwu01/procedural/tie--research-integrity-antipatterns-deceptive-practices-to-reject-in-paper-writing.md
diff --git a/skills/computer-science/machine-learning/ktwu01/procedural/tie--scientific-prose-style-rules-for-avoiding-llm-writing-antipatterns.md b/.agents/skills/computer-science/machine-learning/ktwu01/procedural/tie--scientific-prose-style-rules-for-avoiding-llm-writing-antipatterns.md
similarity index 100%
rename from skills/computer-science/machine-learning/ktwu01/procedural/tie--scientific-prose-style-rules-for-avoiding-llm-writing-antipatterns.md
rename to .agents/skills/computer-science/machine-learning/ktwu01/procedural/tie--scientific-prose-style-rules-for-avoiding-llm-writing-antipatterns.md
diff --git a/skills/computer-science/machine-learning/zhuofli/episodic/failure--diagnosing-degenerate-regression-after-feature-reduction.md b/.agents/skills/computer-science/machine-learning/zhuofli/episodic/failure--diagnosing-degenerate-regression-after-feature-reduction.md
similarity index 100%
rename from skills/computer-science/machine-learning/zhuofli/episodic/failure--diagnosing-degenerate-regression-after-feature-reduction.md
rename to .agents/skills/computer-science/machine-learning/zhuofli/episodic/failure--diagnosing-degenerate-regression-after-feature-reduction.md
diff --git a/skills/computer-science/multiagent-systems/xisen-w/episodic/failure--proof-of-concept-must-implement-paper-algorithm-not-generic-concept.md b/.agents/skills/computer-science/multiagent-systems/xisen-w/episodic/failure--proof-of-concept-must-implement-paper-algorithm-not-generic-concept.md
similarity index 100%
rename from skills/computer-science/multiagent-systems/xisen-w/episodic/failure--proof-of-concept-must-implement-paper-algorithm-not-generic-concept.md
rename to .agents/skills/computer-science/multiagent-systems/xisen-w/episodic/failure--proof-of-concept-must-implement-paper-algorithm-not-generic-concept.md
diff --git a/skills/computer-science/multiagent-systems/xisen-w/procedural/constraint-failure--reframe-comparison-axis-when-caught-in-false-dichotomy.md b/.agents/skills/computer-science/multiagent-systems/xisen-w/procedural/constraint-failure--reframe-comparison-axis-when-caught-in-false-dichotomy.md
similarity index 100%
rename from skills/computer-science/multiagent-systems/xisen-w/procedural/constraint-failure--reframe-comparison-axis-when-caught-in-false-dichotomy.md
rename to .agents/skills/computer-science/multiagent-systems/xisen-w/procedural/constraint-failure--reframe-comparison-axis-when-caught-in-false-dichotomy.md
diff --git a/skills/computer-science/multiagent-systems/xisen-w/procedural/operator-fail--explicit-regime-boundaries-increase-rebuttal-credibility-in-cherry-picked-fields.md b/.agents/skills/computer-science/multiagent-systems/xisen-w/procedural/operator-fail--explicit-regime-boundaries-increase-rebuttal-credibility-in-cherry-picked-fields.md
similarity index 100%
rename from skills/computer-science/multiagent-systems/xisen-w/procedural/operator-fail--explicit-regime-boundaries-increase-rebuttal-credibility-in-cherry-picked-fields.md
rename to .agents/skills/computer-science/multiagent-systems/xisen-w/procedural/operator-fail--explicit-regime-boundaries-increase-rebuttal-credibility-in-cherry-picked-fields.md
diff --git a/skills/computer-science/multiagent-systems/xisen-w/semantic/frontier--llms-as-universal-translators-enable-ontology-free-multi-agent-coordination.md b/.agents/skills/computer-science/multiagent-systems/xisen-w/semantic/frontier--llms-as-universal-translators-enable-ontology-free-multi-agent-coordination.md
similarity index 100%
rename from skills/computer-science/multiagent-systems/xisen-w/semantic/frontier--llms-as-universal-translators-enable-ontology-free-multi-agent-coordination.md
rename to .agents/skills/computer-science/multiagent-systems/xisen-w/semantic/frontier--llms-as-universal-translators-enable-ontology-free-multi-agent-coordination.md
diff --git a/skills/computer-science/software-engineering/anon-b34ee36c/episodic/adaptation--research-memory-extraction-needs-post-pass-curation.md b/.agents/skills/computer-science/software-engineering/anon-b34ee36c/episodic/adaptation--research-memory-extraction-needs-post-pass-curation.md
similarity index 100%
rename from skills/computer-science/software-engineering/anon-b34ee36c/episodic/adaptation--research-memory-extraction-needs-post-pass-curation.md
rename to .agents/skills/computer-science/software-engineering/anon-b34ee36c/episodic/adaptation--research-memory-extraction-needs-post-pass-curation.md
diff --git a/skills/computer-science/software-engineering/anon-b34ee36c/episodic/anomalous--unexpected-template-mixture-in-a-single-site-archive-signals-a-research-data-turning-point.md b/.agents/skills/computer-science/software-engineering/anon-b34ee36c/episodic/anomalous--unexpected-template-mixture-in-a-single-site-archive-signals-a-research-data-turning-point.md
similarity index 100%
rename from skills/computer-science/software-engineering/anon-b34ee36c/episodic/anomalous--unexpected-template-mixture-in-a-single-site-archive-signals-a-research-data-turning-point.md
rename to .agents/skills/computer-science/software-engineering/anon-b34ee36c/episodic/anomalous--unexpected-template-mixture-in-a-single-site-archive-signals-a-research-data-turning-point.md
diff --git a/skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--claim-level-paper-chat-diffing-reveals-tacit-decisions.md b/.agents/skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--claim-level-paper-chat-diffing-reveals-tacit-decisions.md
similarity index 100%
rename from skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--claim-level-paper-chat-diffing-reveals-tacit-decisions.md
rename to .agents/skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--claim-level-paper-chat-diffing-reveals-tacit-decisions.md
diff --git a/skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--predefined-action-vocabulary-with-escape-hatch-for-research-ontology.md b/.agents/skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--predefined-action-vocabulary-with-escape-hatch-for-research-ontology.md
similarity index 100%
rename from skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--predefined-action-vocabulary-with-escape-hatch-for-research-ontology.md
rename to .agents/skills/computer-science/software-engineering/anon-b34ee36c/procedural/tie--predefined-action-vocabulary-with-escape-hatch-for-research-ontology.md
diff --git a/skills/computer-science/software-engineering/anon-b34ee36c/semantic/correction--three-memory-types-for-scientific-knowhow-capture.md b/.agents/skills/computer-science/software-engineering/anon-b34ee36c/semantic/correction--three-memory-types-for-scientific-knowhow-capture.md
similarity index 100%
rename from skills/computer-science/software-engineering/anon-b34ee36c/semantic/correction--three-memory-types-for-scientific-knowhow-capture.md
rename to .agents/skills/computer-science/software-engineering/anon-b34ee36c/semantic/correction--three-memory-types-for-scientific-knowhow-capture.md
diff --git a/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--agent-harness-as-research-workflow-primitives.md b/.agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--agent-harness-as-research-workflow-primitives.md
similarity index 100%
rename from skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--agent-harness-as-research-workflow-primitives.md
rename to .agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--agent-harness-as-research-workflow-primitives.md
diff --git a/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--benchmark-environment-for-ai-agent-scientists.md b/.agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--benchmark-environment-for-ai-agent-scientists.md
similarity index 100%
rename from skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--benchmark-environment-for-ai-agent-scientists.md
rename to .agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--benchmark-environment-for-ai-agent-scientists.md
diff --git a/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--human-authored-skills-registry-for-ai-research-agents.md b/.agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--human-authored-skills-registry-for-ai-research-agents.md
similarity index 100%
rename from skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--human-authored-skills-registry-for-ai-research-agents.md
rename to .agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--human-authored-skills-registry-for-ai-research-agents.md
diff --git a/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--researchskills-skill-memory-taxonomy.md b/.agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--researchskills-skill-memory-taxonomy.md
similarity index 100%
rename from skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--researchskills-skill-memory-taxonomy.md
rename to .agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--researchskills-skill-memory-taxonomy.md
diff --git a/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--skills-fluency-as-scientific-model-objective.md b/.agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--skills-fluency-as-scientific-model-objective.md
similarity index 100%
rename from skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--skills-fluency-as-scientific-model-objective.md
rename to .agents/skills/computer-science/software-engineering/hhhhhejia/semantic/non-public--skills-fluency-as-scientific-model-objective.md
diff --git a/skills/economics/general-economics/clarence-zhewang/semantic/frontier--data-factor-topic-selection-after-field-maturation.md b/.agents/skills/economics/general-economics/clarence-zhewang/semantic/frontier--data-factor-topic-selection-after-field-maturation.md
similarity index 100%
rename from skills/economics/general-economics/clarence-zhewang/semantic/frontier--data-factor-topic-selection-after-field-maturation.md
rename to .agents/skills/economics/general-economics/clarence-zhewang/semantic/frontier--data-factor-topic-selection-after-field-maturation.md
diff --git a/skills/economics/theoretical-economics/clarence-zhewang/episodic/adaptation--transform-parallel-studies-into-build-integrate-optimize-sequence.md b/.agents/skills/economics/theoretical-economics/clarence-zhewang/episodic/adaptation--transform-parallel-studies-into-build-integrate-optimize-sequence.md
similarity index 100%
rename from skills/economics/theoretical-economics/clarence-zhewang/episodic/adaptation--transform-parallel-studies-into-build-integrate-optimize-sequence.md
rename to .agents/skills/economics/theoretical-economics/clarence-zhewang/episodic/adaptation--transform-parallel-studies-into-build-integrate-optimize-sequence.md
diff --git a/skills/management/.gitkeep b/.agents/skills/management/.gitkeep
similarity index 100%
rename from skills/management/.gitkeep
rename to .agents/skills/management/.gitkeep
diff --git a/skills/management/careers/.gitkeep b/.agents/skills/management/careers/.gitkeep
similarity index 100%
rename from skills/management/careers/.gitkeep
rename to .agents/skills/management/careers/.gitkeep
diff --git a/skills/management/communication-digital-technology-and-organization/.gitkeep b/.agents/skills/management/communication-digital-technology-and-organization/.gitkeep
similarity index 100%
rename from skills/management/communication-digital-technology-and-organization/.gitkeep
rename to .agents/skills/management/communication-digital-technology-and-organization/.gitkeep
diff --git a/skills/management/conflict-management/.gitkeep b/.agents/skills/management/conflict-management/.gitkeep
similarity index 100%
rename from skills/management/conflict-management/.gitkeep
rename to .agents/skills/management/conflict-management/.gitkeep
diff --git a/skills/management/critical-management-studies/.gitkeep b/.agents/skills/management/critical-management-studies/.gitkeep
similarity index 100%
rename from skills/management/critical-management-studies/.gitkeep
rename to .agents/skills/management/critical-management-studies/.gitkeep
diff --git a/skills/management/diversity-equity-and-inclusion/.gitkeep b/.agents/skills/management/diversity-equity-and-inclusion/.gitkeep
similarity index 100%
rename from skills/management/diversity-equity-and-inclusion/.gitkeep
rename to .agents/skills/management/diversity-equity-and-inclusion/.gitkeep
diff --git a/skills/management/entrepreneurship/.gitkeep b/.agents/skills/management/entrepreneurship/.gitkeep
similarity index 100%
rename from skills/management/entrepreneurship/.gitkeep
rename to .agents/skills/management/entrepreneurship/.gitkeep
diff --git a/skills/management/health-care-management/.gitkeep b/.agents/skills/management/health-care-management/.gitkeep
similarity index 100%
rename from skills/management/health-care-management/.gitkeep
rename to .agents/skills/management/health-care-management/.gitkeep
diff --git a/skills/management/human-resources/.gitkeep b/.agents/skills/management/human-resources/.gitkeep
similarity index 100%
rename from skills/management/human-resources/.gitkeep
rename to .agents/skills/management/human-resources/.gitkeep
diff --git a/skills/management/international-management/.gitkeep b/.agents/skills/management/international-management/.gitkeep
similarity index 100%
rename from skills/management/international-management/.gitkeep
rename to .agents/skills/management/international-management/.gitkeep
diff --git a/skills/management/management-consulting/.gitkeep b/.agents/skills/management/management-consulting/.gitkeep
similarity index 100%
rename from skills/management/management-consulting/.gitkeep
rename to .agents/skills/management/management-consulting/.gitkeep
diff --git a/skills/management/management-education-and-development/.gitkeep b/.agents/skills/management/management-education-and-development/.gitkeep
similarity index 100%
rename from skills/management/management-education-and-development/.gitkeep
rename to .agents/skills/management/management-education-and-development/.gitkeep
diff --git a/skills/management/management-history/.gitkeep b/.agents/skills/management/management-history/.gitkeep
similarity index 100%
rename from skills/management/management-history/.gitkeep
rename to .agents/skills/management/management-history/.gitkeep
diff --git a/skills/management/management-spirituality-and-religion/.gitkeep b/.agents/skills/management/management-spirituality-and-religion/.gitkeep
similarity index 100%
rename from skills/management/management-spirituality-and-religion/.gitkeep
rename to .agents/skills/management/management-spirituality-and-religion/.gitkeep
diff --git a/skills/management/managerial-and-organizational-cognition/.gitkeep b/.agents/skills/management/managerial-and-organizational-cognition/.gitkeep
similarity index 100%
rename from skills/management/managerial-and-organizational-cognition/.gitkeep
rename to .agents/skills/management/managerial-and-organizational-cognition/.gitkeep
diff --git a/skills/management/operations-and-supply-chain-management/.gitkeep b/.agents/skills/management/operations-and-supply-chain-management/.gitkeep
similarity index 100%
rename from skills/management/operations-and-supply-chain-management/.gitkeep
rename to .agents/skills/management/operations-and-supply-chain-management/.gitkeep
diff --git a/skills/management/organization-and-management-theory/.gitkeep b/.agents/skills/management/organization-and-management-theory/.gitkeep
similarity index 100%
rename from skills/management/organization-and-management-theory/.gitkeep
rename to .agents/skills/management/organization-and-management-theory/.gitkeep
diff --git a/skills/management/organization-development-and-change/.gitkeep b/.agents/skills/management/organization-development-and-change/.gitkeep
similarity index 100%
rename from skills/management/organization-development-and-change/.gitkeep
rename to .agents/skills/management/organization-development-and-change/.gitkeep
diff --git a/skills/management/organizational-behavior/.gitkeep b/.agents/skills/management/organizational-behavior/.gitkeep
similarity index 100%
rename from skills/management/organizational-behavior/.gitkeep
rename to .agents/skills/management/organizational-behavior/.gitkeep
diff --git a/skills/management/organizational-neuroscience-and-biology/.gitkeep b/.agents/skills/management/organizational-neuroscience-and-biology/.gitkeep
similarity index 100%
rename from skills/management/organizational-neuroscience-and-biology/.gitkeep
rename to .agents/skills/management/organizational-neuroscience-and-biology/.gitkeep
diff --git a/skills/management/organizations-and-the-natural-environment/.gitkeep b/.agents/skills/management/organizations-and-the-natural-environment/.gitkeep
similarity index 100%
rename from skills/management/organizations-and-the-natural-environment/.gitkeep
rename to .agents/skills/management/organizations-and-the-natural-environment/.gitkeep
diff --git a/skills/management/public-and-nonprofit/.gitkeep b/.agents/skills/management/public-and-nonprofit/.gitkeep
similarity index 100%
rename from skills/management/public-and-nonprofit/.gitkeep
rename to .agents/skills/management/public-and-nonprofit/.gitkeep
diff --git a/skills/management/research-methods/.gitkeep b/.agents/skills/management/research-methods/.gitkeep
similarity index 100%
rename from skills/management/research-methods/.gitkeep
rename to .agents/skills/management/research-methods/.gitkeep
diff --git a/skills/management/social-issues-in-management/.gitkeep b/.agents/skills/management/social-issues-in-management/.gitkeep
similarity index 100%
rename from skills/management/social-issues-in-management/.gitkeep
rename to .agents/skills/management/social-issues-in-management/.gitkeep
diff --git a/skills/management/strategic-management/.gitkeep b/.agents/skills/management/strategic-management/.gitkeep
similarity index 100%
rename from skills/management/strategic-management/.gitkeep
rename to .agents/skills/management/strategic-management/.gitkeep
diff --git a/skills/management/strategizing-activities-and-practices/.gitkeep b/.agents/skills/management/strategizing-activities-and-practices/.gitkeep
similarity index 100%
rename from skills/management/strategizing-activities-and-practices/.gitkeep
rename to .agents/skills/management/strategizing-activities-and-practices/.gitkeep
diff --git a/skills/management/technology-and-innovation-management/.gitkeep b/.agents/skills/management/technology-and-innovation-management/.gitkeep
similarity index 100%
rename from skills/management/technology-and-innovation-management/.gitkeep
rename to .agents/skills/management/technology-and-innovation-management/.gitkeep
diff --git a/skills/mathematics/approximation-theory/pillow0705/semantic/frontier--polynomial-approximation-of-neural-networks-achieves-16x-inference-speedup-via-weierstrass-construction.md b/.agents/skills/mathematics/approximation-theory/pillow0705/semantic/frontier--polynomial-approximation-of-neural-networks-achieves-16x-inference-speedup-via-weierstrass-construction.md
similarity index 100%
rename from skills/mathematics/approximation-theory/pillow0705/semantic/frontier--polynomial-approximation-of-neural-networks-achieves-16x-inference-speedup-via-weierstrass-construction.md
rename to .agents/skills/mathematics/approximation-theory/pillow0705/semantic/frontier--polynomial-approximation-of-neural-networks-achieves-16x-inference-speedup-via-weierstrass-construction.md
diff --git a/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-afternoon-psn-depression-vs-shs-enhancement.md b/.agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-afternoon-psn-depression-vs-shs-enhancement.md
similarity index 100%
rename from skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-afternoon-psn-depression-vs-shs-enhancement.md
rename to .agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-afternoon-psn-depression-vs-shs-enhancement.md
diff --git a/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-subtropical-degradation-28-32n.md b/.agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-subtropical-degradation-28-32n.md
similarity index 100%
rename from skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-subtropical-degradation-28-32n.md
rename to .agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-subtropical-degradation-28-32n.md
diff --git a/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-two-pathway-transpiration-violates-standard-water-balance-check.md b/.agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-two-pathway-transpiration-violates-standard-water-balance-check.md
similarity index 100%
rename from skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-two-pathway-transpiration-violates-standard-water-balance-check.md
rename to .agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/anomalous--phs-two-pathway-transpiration-violates-standard-water-balance-check.md
diff --git a/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/failure--falsifiability-as-minimum-bar-for-scientific-claims.md b/.agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/failure--falsifiability-as-minimum-bar-for-scientific-claims.md
similarity index 100%
rename from skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/failure--falsifiability-as-minimum-bar-for-scientific-claims.md
rename to .agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/episodic/failure--falsifiability-as-minimum-bar-for-scientific-claims.md
diff --git a/skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--legacy-phs-v3-conus-domain-invalidity.md b/.agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--legacy-phs-v3-conus-domain-invalidity.md
similarity index 100%
rename from skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--legacy-phs-v3-conus-domain-invalidity.md
rename to .agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--legacy-phs-v3-conus-domain-invalidity.md
diff --git a/skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--phs-vpsis-physical-clamp-at-minus-9000mm.md b/.agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--phs-vpsis-physical-clamp-at-minus-9000mm.md
similarity index 100%
rename from skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--phs-vpsis-physical-clamp-at-minus-9000mm.md
rename to .agents/skills/physics/atmospheric-and-oceanic-physics/ktwu01/semantic/correction--phs-vpsis-physical-clamp-at-minus-9000mm.md
diff --git a/skills/physics/computational-physics/huangzesen/procedural/constraint-failure--dealias-before-helmholtz-divergence-removal.md b/.agents/skills/physics/computational-physics/huangzesen/procedural/constraint-failure--dealias-before-helmholtz-divergence-removal.md
similarity index 100%
rename from skills/physics/computational-physics/huangzesen/procedural/constraint-failure--dealias-before-helmholtz-divergence-removal.md
rename to .agents/skills/physics/computational-physics/huangzesen/procedural/constraint-failure--dealias-before-helmholtz-divergence-removal.md
diff --git a/skills/physics/computational-physics/huangzesen/semantic/correction--helmholtz-projection-nyquist-residual.md b/.agents/skills/physics/computational-physics/huangzesen/semantic/correction--helmholtz-projection-nyquist-residual.md
similarity index 100%
rename from skills/physics/computational-physics/huangzesen/semantic/correction--helmholtz-projection-nyquist-residual.md
rename to .agents/skills/physics/computational-physics/huangzesen/semantic/correction--helmholtz-projection-nyquist-residual.md
diff --git a/skills/physics/computational-physics/huangzesen/semantic/non-public--laps-mhd-output-float-precision.md b/.agents/skills/physics/computational-physics/huangzesen/semantic/non-public--laps-mhd-output-float-precision.md
similarity index 100%
rename from skills/physics/computational-physics/huangzesen/semantic/non-public--laps-mhd-output-float-precision.md
rename to .agents/skills/physics/computational-physics/huangzesen/semantic/non-public--laps-mhd-output-float-precision.md
diff --git a/skills/physics/computational-physics/pillow0705/episodic/anomalous--absolute-collision-tolerance-breaks-scale-invariance-in-sphere-packing-simulations.md b/.agents/skills/physics/computational-physics/pillow0705/episodic/anomalous--absolute-collision-tolerance-breaks-scale-invariance-in-sphere-packing-simulations.md
similarity index 100%
rename from skills/physics/computational-physics/pillow0705/episodic/anomalous--absolute-collision-tolerance-breaks-scale-invariance-in-sphere-packing-simulations.md
rename to .agents/skills/physics/computational-physics/pillow0705/episodic/anomalous--absolute-collision-tolerance-breaks-scale-invariance-in-sphere-packing-simulations.md
diff --git a/skills/physics/computational-physics/pillow0705/procedural/tie--particle-size-distribution-shape-dominates-model-architecture-in-rl-packing-policy.md b/.agents/skills/physics/computational-physics/pillow0705/procedural/tie--particle-size-distribution-shape-dominates-model-architecture-in-rl-packing-policy.md
similarity index 100%
rename from skills/physics/computational-physics/pillow0705/procedural/tie--particle-size-distribution-shape-dominates-model-architecture-in-rl-packing-policy.md
rename to .agents/skills/physics/computational-physics/pillow0705/procedural/tie--particle-size-distribution-shape-dominates-model-architecture-in-rl-packing-policy.md
diff --git a/skills/physics/computational-physics/pillow0705/semantic/frontier--embedding-ratio-threshold-determines-packing-regime-in-polydisperse-sphere-packing.md b/.agents/skills/physics/computational-physics/pillow0705/semantic/frontier--embedding-ratio-threshold-determines-packing-regime-in-polydisperse-sphere-packing.md
similarity index 100%
rename from skills/physics/computational-physics/pillow0705/semantic/frontier--embedding-ratio-threshold-determines-packing-regime-in-polydisperse-sphere-packing.md
rename to .agents/skills/physics/computational-physics/pillow0705/semantic/frontier--embedding-ratio-threshold-determines-packing-regime-in-polydisperse-sphere-packing.md
diff --git a/skills/physics/computational-physics/pillow0705/semantic/frontier--extended-training-runs-degrade-eval-performance-in-reinforce-based-packing-policies.md b/.agents/skills/physics/computational-physics/pillow0705/semantic/frontier--extended-training-runs-degrade-eval-performance-in-reinforce-based-packing-policies.md
similarity index 100%
rename from skills/physics/computational-physics/pillow0705/semantic/frontier--extended-training-runs-degrade-eval-performance-in-reinforce-based-packing-policies.md
rename to .agents/skills/physics/computational-physics/pillow0705/semantic/frontier--extended-training-runs-degrade-eval-performance-in-reinforce-based-packing-policies.md
diff --git a/skills/physics/computational-physics/pillow0705/semantic/frontier--three-sphere-tangency-generates-physically-valid-candidates-analytically-in-sequential-packing.md b/.agents/skills/physics/computational-physics/pillow0705/semantic/frontier--three-sphere-tangency-generates-physically-valid-candidates-analytically-in-sequential-packing.md
similarity index 100%
rename from skills/physics/computational-physics/pillow0705/semantic/frontier--three-sphere-tangency-generates-physically-valid-candidates-analytically-in-sequential-packing.md
rename to .agents/skills/physics/computational-physics/pillow0705/semantic/frontier--three-sphere-tangency-generates-physically-valid-candidates-analytically-in-sequential-packing.md
diff --git a/skills/physics/computational-physics/pillow0705/semantic/frontier--two-stage-collapse-model-predicts-training-failure-in-rl-sphere-packing.md b/.agents/skills/physics/computational-physics/pillow0705/semantic/frontier--two-stage-collapse-model-predicts-training-failure-in-rl-sphere-packing.md
similarity index 100%
rename from skills/physics/computational-physics/pillow0705/semantic/frontier--two-stage-collapse-model-predicts-training-failure-in-rl-sphere-packing.md
rename to .agents/skills/physics/computational-physics/pillow0705/semantic/frontier--two-stage-collapse-model-predicts-training-failure-in-rl-sphere-packing.md
diff --git a/skills/physics/geophysics/ktwu01/episodic/adaptation--apply-the-explicit-review-rubric-to-the-manuscript-not-to-the-rubric-itself.md b/.agents/skills/physics/geophysics/ktwu01/episodic/adaptation--apply-the-explicit-review-rubric-to-the-manuscript-not-to-the-rubric-itself.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/episodic/adaptation--apply-the-explicit-review-rubric-to-the-manuscript-not-to-the-rubric-itself.md
rename to .agents/skills/physics/geophysics/ktwu01/episodic/adaptation--apply-the-explicit-review-rubric-to-the-manuscript-not-to-the-rubric-itself.md
diff --git a/skills/physics/geophysics/ktwu01/episodic/adaptation--benchmark-and-frontier-gap-scan-for-review-articles.md b/.agents/skills/physics/geophysics/ktwu01/episodic/adaptation--benchmark-and-frontier-gap-scan-for-review-articles.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/episodic/adaptation--benchmark-and-frontier-gap-scan-for-review-articles.md
rename to .agents/skills/physics/geophysics/ktwu01/episodic/adaptation--benchmark-and-frontier-gap-scan-for-review-articles.md
diff --git a/skills/physics/geophysics/ktwu01/episodic/adaptation--hackathon-interactive-webapp-as-differentiator.md b/.agents/skills/physics/geophysics/ktwu01/episodic/adaptation--hackathon-interactive-webapp-as-differentiator.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/episodic/adaptation--hackathon-interactive-webapp-as-differentiator.md
rename to .agents/skills/physics/geophysics/ktwu01/episodic/adaptation--hackathon-interactive-webapp-as-differentiator.md
diff --git a/skills/physics/geophysics/ktwu01/episodic/adaptation--hysteresis-scanning-curve-validation.md b/.agents/skills/physics/geophysics/ktwu01/episodic/adaptation--hysteresis-scanning-curve-validation.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/episodic/adaptation--hysteresis-scanning-curve-validation.md
rename to .agents/skills/physics/geophysics/ktwu01/episodic/adaptation--hysteresis-scanning-curve-validation.md
diff --git a/skills/physics/geophysics/ktwu01/episodic/anomalous--detecting-structural-model-mismatch-via-seasonal-dynamics.md b/.agents/skills/physics/geophysics/ktwu01/episodic/anomalous--detecting-structural-model-mismatch-via-seasonal-dynamics.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/episodic/anomalous--detecting-structural-model-mismatch-via-seasonal-dynamics.md
rename to .agents/skills/physics/geophysics/ktwu01/episodic/anomalous--detecting-structural-model-mismatch-via-seasonal-dynamics.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/constraint-failure--grace-resolution-vs-rock-moisture-scale-mismatch.md b/.agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--grace-resolution-vs-rock-moisture-scale-mismatch.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/constraint-failure--grace-resolution-vs-rock-moisture-scale-mismatch.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--grace-resolution-vs-rock-moisture-scale-mismatch.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/constraint-failure--hydrology-model-integration-of-rock-moisture.md b/.agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--hydrology-model-integration-of-rock-moisture.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/constraint-failure--hydrology-model-integration-of-rock-moisture.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--hydrology-model-integration-of-rock-moisture.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/constraint-failure--llm-agent-fortran-refactoring.md b/.agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--llm-agent-fortran-refactoring.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/constraint-failure--llm-agent-fortran-refactoring.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--llm-agent-fortran-refactoring.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/constraint-failure--ocean-color-vs-land-surface-products-for-coastal-turbidity.md b/.agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--ocean-color-vs-land-surface-products-for-coastal-turbidity.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/constraint-failure--ocean-color-vs-land-surface-products-for-coastal-turbidity.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--ocean-color-vs-land-surface-products-for-coastal-turbidity.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/constraint-failure--reviewer-flag-unnamed-datasets-require-named-citations.md b/.agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--reviewer-flag-unnamed-datasets-require-named-citations.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/constraint-failure--reviewer-flag-unnamed-datasets-require-named-citations.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--reviewer-flag-unnamed-datasets-require-named-citations.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/constraint-failure--sap-flow-upscaling-uncertainty-into-likelihood.md b/.agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--sap-flow-upscaling-uncertainty-into-likelihood.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/constraint-failure--sap-flow-upscaling-uncertainty-into-likelihood.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/constraint-failure--sap-flow-upscaling-uncertainty-into-likelihood.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/no-change--diagnosing-overconfidence-induced-research-stall.md b/.agents/skills/physics/geophysics/ktwu01/procedural/no-change--diagnosing-overconfidence-induced-research-stall.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/no-change--diagnosing-overconfidence-induced-research-stall.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/no-change--diagnosing-overconfidence-induced-research-stall.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/no-change--enforcing-output-over-tool-optimization-in-research-systems.md b/.agents/skills/physics/geophysics/ktwu01/procedural/no-change--enforcing-output-over-tool-optimization-in-research-systems.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/no-change--enforcing-output-over-tool-optimization-in-research-systems.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/no-change--enforcing-output-over-tool-optimization-in-research-systems.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/no-change--forcing-research-decomposition-under-vague-problem-definition.md b/.agents/skills/physics/geophysics/ktwu01/procedural/no-change--forcing-research-decomposition-under-vague-problem-definition.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/no-change--forcing-research-decomposition-under-vague-problem-definition.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/no-change--forcing-research-decomposition-under-vague-problem-definition.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/operator-fail--cross-cutting-validation-four-variable-types.md b/.agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--cross-cutting-validation-four-variable-types.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/operator-fail--cross-cutting-validation-four-variable-types.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--cross-cutting-validation-four-variable-types.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/operator-fail--dual-porosity-convergence-tuning.md b/.agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--dual-porosity-convergence-tuning.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/operator-fail--dual-porosity-convergence-tuning.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--dual-porosity-convergence-tuning.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/operator-fail--ensemble-kalman-filter-smap-grace-localization.md b/.agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--ensemble-kalman-filter-smap-grace-localization.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/operator-fail--ensemble-kalman-filter-smap-grace-localization.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--ensemble-kalman-filter-smap-grace-localization.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/operator-fail--factorial-experiments-to-isolate-rock-vs-groundwater-contributions.md b/.agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--factorial-experiments-to-isolate-rock-vs-groundwater-contributions.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/operator-fail--factorial-experiments-to-isolate-rock-vs-groundwater-contributions.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--factorial-experiments-to-isolate-rock-vs-groundwater-contributions.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/operator-fail--mualem-independent-domain-path-dependent-psi-stem.md b/.agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--mualem-independent-domain-path-dependent-psi-stem.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/operator-fail--mualem-independent-domain-path-dependent-psi-stem.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--mualem-independent-domain-path-dependent-psi-stem.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/operator-fail--three-tier-falsifiability-success-criteria-rock-moisture.md b/.agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--three-tier-falsifiability-success-criteria-rock-moisture.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/operator-fail--three-tier-falsifiability-success-criteria-rock-moisture.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/operator-fail--three-tier-falsifiability-success-criteria-rock-moisture.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/tie--distinguishing-parameter-error-vs-structural-error-in-models.md b/.agents/skills/physics/geophysics/ktwu01/procedural/tie--distinguishing-parameter-error-vs-structural-error-in-models.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/tie--distinguishing-parameter-error-vs-structural-error-in-models.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/tie--distinguishing-parameter-error-vs-structural-error-in-models.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/tie--exploiting-cross-domain-concepts-for-hydrological-innovation.md b/.agents/skills/physics/geophysics/ktwu01/procedural/tie--exploiting-cross-domain-concepts-for-hydrological-innovation.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/tie--exploiting-cross-domain-concepts-for-hydrological-innovation.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/tie--exploiting-cross-domain-concepts-for-hydrological-innovation.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/tie--gmd-vs-james-vs-ai-venue-for-ai-agents-for-esm.md b/.agents/skills/physics/geophysics/ktwu01/procedural/tie--gmd-vs-james-vs-ai-venue-for-ai-agents-for-esm.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/tie--gmd-vs-james-vs-ai-venue-for-ai-agents-for-esm.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/tie--gmd-vs-james-vs-ai-venue-for-ai-agents-for-esm.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/tie--lightweight-classifier-over-finetuning-for-foundation-embeddings.md b/.agents/skills/physics/geophysics/ktwu01/procedural/tie--lightweight-classifier-over-finetuning-for-foundation-embeddings.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/tie--lightweight-classifier-over-finetuning-for-foundation-embeddings.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/tie--lightweight-classifier-over-finetuning-for-foundation-embeddings.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/tie--prioritizing-seasonal-patterns-over-pointwise-fit-in-geoscience.md b/.agents/skills/physics/geophysics/ktwu01/procedural/tie--prioritizing-seasonal-patterns-over-pointwise-fit-in-geoscience.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/tie--prioritizing-seasonal-patterns-over-pointwise-fit-in-geoscience.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/tie--prioritizing-seasonal-patterns-over-pointwise-fit-in-geoscience.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/tie--sce-ua-vs-mcmc-dream-for-hydraulic-parameter-calibration.md b/.agents/skills/physics/geophysics/ktwu01/procedural/tie--sce-ua-vs-mcmc-dream-for-hydraulic-parameter-calibration.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/tie--sce-ua-vs-mcmc-dream-for-hydraulic-parameter-calibration.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/tie--sce-ua-vs-mcmc-dream-for-hydraulic-parameter-calibration.md
diff --git a/skills/physics/geophysics/ktwu01/procedural/tie--sequential-calibration-to-manage-parameter-compounding.md b/.agents/skills/physics/geophysics/ktwu01/procedural/tie--sequential-calibration-to-manage-parameter-compounding.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/procedural/tie--sequential-calibration-to-manage-parameter-compounding.md
rename to .agents/skills/physics/geophysics/ktwu01/procedural/tie--sequential-calibration-to-manage-parameter-compounding.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--bayesian-parameter-estimation-is-data-assimilation-naming.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--bayesian-parameter-estimation-is-data-assimilation-naming.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--bayesian-parameter-estimation-is-data-assimilation-naming.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--bayesian-parameter-estimation-is-data-assimilation-naming.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--benchmarking-is-a-core-axis-not-a-side-note-in-modern-lsm-reviews.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--benchmarking-is-a-core-axis-not-a-side-note-in-modern-lsm-reviews.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--benchmarking-is-a-core-axis-not-a-side-note-in-modern-lsm-reviews.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--benchmarking-is-a-core-axis-not-a-side-note-in-modern-lsm-reviews.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--blanket-superiority-claims-for-ml-in-environmental-modelling-need-benchmark-and-mechanism-qualifiers.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--blanket-superiority-claims-for-ml-in-environmental-modelling-need-benchmark-and-mechanism-qualifiers.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--blanket-superiority-claims-for-ml-in-environmental-modelling-need-benchmark-and-mechanism-qualifiers.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--blanket-superiority-claims-for-ml-in-environmental-modelling-need-benchmark-and-mechanism-qualifiers.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--hugo-fabricated-sapwood-50-percent-daily-transpiration.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--hugo-fabricated-sapwood-50-percent-daily-transpiration.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--hugo-fabricated-sapwood-50-percent-daily-transpiration.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--hugo-fabricated-sapwood-50-percent-daily-transpiration.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--lsm-grid-resolution-is-a-design-choice-not-a-physical-limit.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--lsm-grid-resolution-is-a-design-choice-not-a-physical-limit.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--lsm-grid-resolution-is-a-design-choice-not-a-physical-limit.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--lsm-grid-resolution-is-a-design-choice-not-a-physical-limit.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--noah-mp-wood-moisture-initialization.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--noah-mp-wood-moisture-initialization.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--noah-mp-wood-moisture-initialization.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--noah-mp-wood-moisture-initialization.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--sif-light-vs-dark-reactions-decoupling-under-stress.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--sif-light-vs-dark-reactions-decoupling-under-stress.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--sif-light-vs-dark-reactions-decoupling-under-stress.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--sif-light-vs-dark-reactions-decoupling-under-stress.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--split-sample-vs-leave-one-out-cross-validation-distinction.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--split-sample-vs-leave-one-out-cross-validation-distinction.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--split-sample-vs-leave-one-out-cross-validation-distinction.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--split-sample-vs-leave-one-out-cross-validation-distinction.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/correction--temporal-axis-for-lsm-domain-figures-correct-semantic-interpretation.md b/.agents/skills/physics/geophysics/ktwu01/semantic/correction--temporal-axis-for-lsm-domain-figures-correct-semantic-interpretation.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/correction--temporal-axis-for-lsm-domain-figures-correct-semantic-interpretation.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/correction--temporal-axis-for-lsm-domain-figures-correct-semantic-interpretation.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/frontier--alphaearth-10m-64band-annual-embeddings.md b/.agents/skills/physics/geophysics/ktwu01/semantic/frontier--alphaearth-10m-64band-annual-embeddings.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/frontier--alphaearth-10m-64band-annual-embeddings.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/frontier--alphaearth-10m-64band-annual-embeddings.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/frontier--deficit-based-mass-balance-lower-bound-sbedrock.md b/.agents/skills/physics/geophysics/ktwu01/semantic/frontier--deficit-based-mass-balance-lower-bound-sbedrock.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/frontier--deficit-based-mass-balance-lower-bound-sbedrock.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/frontier--deficit-based-mass-balance-lower-bound-sbedrock.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/frontier--dual-anonymous-peer-review-formatting-finesst.md b/.agents/skills/physics/geophysics/ktwu01/semantic/frontier--dual-anonymous-peer-review-formatting-finesst.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/frontier--dual-anonymous-peer-review-formatting-finesst.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/frontier--dual-anonymous-peer-review-formatting-finesst.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/frontier--mcp-physics-models-novelty-gap.md b/.agents/skills/physics/geophysics/ktwu01/semantic/frontier--mcp-physics-models-novelty-gap.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/frontier--mcp-physics-models-novelty-gap.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/frontier--mcp-physics-models-novelty-gap.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/frontier--meteor-detection-radon-transform-vs-cnn-pipeline.md b/.agents/skills/physics/geophysics/ktwu01/semantic/frontier--meteor-detection-radon-transform-vs-cnn-pipeline.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/frontier--meteor-detection-radon-transform-vs-cnn-pipeline.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/frontier--meteor-detection-radon-transform-vs-cnn-pipeline.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/frontier--rock-moisture-storage-capacity-scaling.md b/.agents/skills/physics/geophysics/ktwu01/semantic/frontier--rock-moisture-storage-capacity-scaling.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/frontier--rock-moisture-storage-capacity-scaling.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/frontier--rock-moisture-storage-capacity-scaling.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/frontier--van-genuchten-hysteresis-stem-water-curve.md b/.agents/skills/physics/geophysics/ktwu01/semantic/frontier--van-genuchten-hysteresis-stem-water-curve.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/frontier--van-genuchten-hysteresis-stem-water-curve.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/frontier--van-genuchten-hysteresis-stem-water-curve.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/non-public--hunga-tonga-lamb-wave-two-mode-propagation.md b/.agents/skills/physics/geophysics/ktwu01/semantic/non-public--hunga-tonga-lamb-wave-two-mode-propagation.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/non-public--hunga-tonga-lamb-wave-two-mode-propagation.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/non-public--hunga-tonga-lamb-wave-two-mode-propagation.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/non-public--mengcheng-meteor-radar-double-gaussian-dual-peak.md b/.agents/skills/physics/geophysics/ktwu01/semantic/non-public--mengcheng-meteor-radar-double-gaussian-dual-peak.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/non-public--mengcheng-meteor-radar-double-gaussian-dual-peak.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/non-public--mengcheng-meteor-radar-double-gaussian-dual-peak.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-dual-porosity-bedrock-extension-spec.md b/.agents/skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-dual-porosity-bedrock-extension-spec.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-dual-porosity-bedrock-extension-spec.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-dual-porosity-bedrock-extension-spec.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-ncar-hpc-allocation-cpu-budget.md b/.agents/skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-ncar-hpc-allocation-cpu-budget.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-ncar-hpc-allocation-cpu-budget.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/non-public--noah-mp-ncar-hpc-allocation-cpu-budget.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/non-public--pelletier-regolith-thickness-dataset-for-noahmp-depth.md b/.agents/skills/physics/geophysics/ktwu01/semantic/non-public--pelletier-regolith-thickness-dataset-for-noahmp-depth.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/non-public--pelletier-regolith-thickness-dataset-for-noahmp-depth.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/non-public--pelletier-regolith-thickness-dataset-for-noahmp-depth.md
diff --git a/skills/physics/geophysics/ktwu01/semantic/non-public--phase-1-prior-work-line-counts-fortran.md b/.agents/skills/physics/geophysics/ktwu01/semantic/non-public--phase-1-prior-work-line-counts-fortran.md
similarity index 100%
rename from skills/physics/geophysics/ktwu01/semantic/non-public--phase-1-prior-work-line-counts-fortran.md
rename to .agents/skills/physics/geophysics/ktwu01/semantic/non-public--phase-1-prior-work-line-counts-fortran.md
diff --git a/skills/physics/medical-physics/wangmengxiao319/procedural/tie--varp-checkpoint-reuse-to-skip-s1-re-simulation.md b/.agents/skills/physics/medical-physics/wangmengxiao319/procedural/tie--varp-checkpoint-reuse-to-skip-s1-re-simulation.md
similarity index 100%
rename from skills/physics/medical-physics/wangmengxiao319/procedural/tie--varp-checkpoint-reuse-to-skip-s1-re-simulation.md
rename to .agents/skills/physics/medical-physics/wangmengxiao319/procedural/tie--varp-checkpoint-reuse-to-skip-s1-re-simulation.md
diff --git a/skills/physics/medical-physics/wangmengxiao319/semantic/correction--cardiac-tissue-limit-cycle-is-pacing-site-specific.md b/.agents/skills/physics/medical-physics/wangmengxiao319/semantic/correction--cardiac-tissue-limit-cycle-is-pacing-site-specific.md
similarity index 100%
rename from skills/physics/medical-physics/wangmengxiao319/semantic/correction--cardiac-tissue-limit-cycle-is-pacing-site-specific.md
rename to .agents/skills/physics/medical-physics/wangmengxiao319/semantic/correction--cardiac-tissue-limit-cycle-is-pacing-site-specific.md
diff --git a/skills/physics/medical-physics/wangmengxiao319/semantic/correction--ventricular-varp-pacing-sites-are-endocardial-not-epicardial.md b/.agents/skills/physics/medical-physics/wangmengxiao319/semantic/correction--ventricular-varp-pacing-sites-are-endocardial-not-epicardial.md
similarity index 100%
rename from skills/physics/medical-physics/wangmengxiao319/semantic/correction--ventricular-varp-pacing-sites-are-endocardial-not-epicardial.md
rename to .agents/skills/physics/medical-physics/wangmengxiao319/semantic/correction--ventricular-varp-pacing-sites-are-endocardial-not-epicardial.md
diff --git a/skills/physics/space-physics/kkeerb/episodic/adaptation--ssj-conductance-first-visualization-over-raw-spectra.md b/.agents/skills/physics/space-physics/kkeerb/episodic/adaptation--ssj-conductance-first-visualization-over-raw-spectra.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/episodic/adaptation--ssj-conductance-first-visualization-over-raw-spectra.md
rename to .agents/skills/physics/space-physics/kkeerb/episodic/adaptation--ssj-conductance-first-visualization-over-raw-spectra.md
diff --git a/skills/physics/space-physics/kkeerb/episodic/anomalous--fte-satellite-relative-motion-reference-frame.md b/.agents/skills/physics/space-physics/kkeerb/episodic/anomalous--fte-satellite-relative-motion-reference-frame.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/episodic/anomalous--fte-satellite-relative-motion-reference-frame.md
rename to .agents/skills/physics/space-physics/kkeerb/episodic/anomalous--fte-satellite-relative-motion-reference-frame.md
diff --git a/skills/physics/space-physics/kkeerb/episodic/failure--fte-satellite-reference-frame-fte-advection-dominates-spacecraft-motion.md b/.agents/skills/physics/space-physics/kkeerb/episodic/failure--fte-satellite-reference-frame-fte-advection-dominates-spacecraft-motion.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/episodic/failure--fte-satellite-reference-frame-fte-advection-dominates-spacecraft-motion.md
rename to .agents/skills/physics/space-physics/kkeerb/episodic/failure--fte-satellite-reference-frame-fte-advection-dominates-spacecraft-motion.md
diff --git a/skills/physics/space-physics/kkeerb/procedural/constraint-failure--bipolar-bn-as-sampling-artifact-of-moving-structure.md b/.agents/skills/physics/space-physics/kkeerb/procedural/constraint-failure--bipolar-bn-as-sampling-artifact-of-moving-structure.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/procedural/constraint-failure--bipolar-bn-as-sampling-artifact-of-moving-structure.md
rename to .agents/skills/physics/space-physics/kkeerb/procedural/constraint-failure--bipolar-bn-as-sampling-artifact-of-moving-structure.md
diff --git a/skills/physics/space-physics/kkeerb/semantic/correction--bn-0-contour-is-not-the-flux-rope-cross-section.md b/.agents/skills/physics/space-physics/kkeerb/semantic/correction--bn-0-contour-is-not-the-flux-rope-cross-section.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/semantic/correction--bn-0-contour-is-not-the-flux-rope-cross-section.md
rename to .agents/skills/physics/space-physics/kkeerb/semantic/correction--bn-0-contour-is-not-the-flux-rope-cross-section.md
diff --git a/skills/physics/space-physics/kkeerb/semantic/correction--bn-0-is-a-projected-identification-proxy-not-a-true-3d-magnetic-axis.md b/.agents/skills/physics/space-physics/kkeerb/semantic/correction--bn-0-is-a-projected-identification-proxy-not-a-true-3d-magnetic-axis.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/semantic/correction--bn-0-is-a-projected-identification-proxy-not-a-true-3d-magnetic-axis.md
rename to .agents/skills/physics/space-physics/kkeerb/semantic/correction--bn-0-is-a-projected-identification-proxy-not-a-true-3d-magnetic-axis.md
diff --git a/skills/physics/space-physics/kkeerb/semantic/correction--robinson-1987-conductance-formula-full-vs-simplified-form-discrepancy.md b/.agents/skills/physics/space-physics/kkeerb/semantic/correction--robinson-1987-conductance-formula-full-vs-simplified-form-discrepancy.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/semantic/correction--robinson-1987-conductance-formula-full-vs-simplified-form-discrepancy.md
rename to .agents/skills/physics/space-physics/kkeerb/semantic/correction--robinson-1987-conductance-formula-full-vs-simplified-form-discrepancy.md
diff --git a/skills/physics/space-physics/kkeerb/semantic/correction--robinson-formula-yields-height-integrated-conductance-not-local-conductivity.md b/.agents/skills/physics/space-physics/kkeerb/semantic/correction--robinson-formula-yields-height-integrated-conductance-not-local-conductivity.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/semantic/correction--robinson-formula-yields-height-integrated-conductance-not-local-conductivity.md
rename to .agents/skills/physics/space-physics/kkeerb/semantic/correction--robinson-formula-yields-height-integrated-conductance-not-local-conductivity.md
diff --git a/skills/physics/space-physics/kkeerb/semantic/frontier--instrument-provenance-determines-data-selection-in-multi-instrument-reproduction.md b/.agents/skills/physics/space-physics/kkeerb/semantic/frontier--instrument-provenance-determines-data-selection-in-multi-instrument-reproduction.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/semantic/frontier--instrument-provenance-determines-data-selection-in-multi-instrument-reproduction.md
rename to .agents/skills/physics/space-physics/kkeerb/semantic/frontier--instrument-provenance-determines-data-selection-in-multi-instrument-reproduction.md
diff --git a/skills/physics/space-physics/kkeerb/semantic/frontier--two-layer-data-source-model-for-satellite-plotting-tools.md b/.agents/skills/physics/space-physics/kkeerb/semantic/frontier--two-layer-data-source-model-for-satellite-plotting-tools.md
similarity index 100%
rename from skills/physics/space-physics/kkeerb/semantic/frontier--two-layer-data-source-model-for-satellite-plotting-tools.md
rename to .agents/skills/physics/space-physics/kkeerb/semantic/frontier--two-layer-data-source-model-for-satellite-plotting-tools.md
diff --git a/skills/quantitative-biology/genomics/ergan-shang/episodic/adaptation--trvae-unpaired-condition-generation.md b/.agents/skills/quantitative-biology/genomics/ergan-shang/episodic/adaptation--trvae-unpaired-condition-generation.md
similarity index 100%
rename from skills/quantitative-biology/genomics/ergan-shang/episodic/adaptation--trvae-unpaired-condition-generation.md
rename to .agents/skills/quantitative-biology/genomics/ergan-shang/episodic/adaptation--trvae-unpaired-condition-generation.md
diff --git a/skills/quantitative-biology/genomics/ergan-shang/procedural/constraint-failure--slaf-large-scale-anndata-merging.md b/.agents/skills/quantitative-biology/genomics/ergan-shang/procedural/constraint-failure--slaf-large-scale-anndata-merging.md
similarity index 100%
rename from skills/quantitative-biology/genomics/ergan-shang/procedural/constraint-failure--slaf-large-scale-anndata-merging.md
rename to .agents/skills/quantitative-biology/genomics/ergan-shang/procedural/constraint-failure--slaf-large-scale-anndata-merging.md
diff --git a/skills/quantitative-biology/genomics/jamesyu420/procedural/operator-fail--permutation-invariant-program-recovery-protocol.md b/.agents/skills/quantitative-biology/genomics/jamesyu420/procedural/operator-fail--permutation-invariant-program-recovery-protocol.md
similarity index 100%
rename from skills/quantitative-biology/genomics/jamesyu420/procedural/operator-fail--permutation-invariant-program-recovery-protocol.md
rename to .agents/skills/quantitative-biology/genomics/jamesyu420/procedural/operator-fail--permutation-invariant-program-recovery-protocol.md
diff --git a/skills/quantitative-biology/genomics/jamesyu420/procedural/tie--graph-guided-spatial-programs-need-uq.md b/.agents/skills/quantitative-biology/genomics/jamesyu420/procedural/tie--graph-guided-spatial-programs-need-uq.md
similarity index 100%
rename from skills/quantitative-biology/genomics/jamesyu420/procedural/tie--graph-guided-spatial-programs-need-uq.md
rename to .agents/skills/quantitative-biology/genomics/jamesyu420/procedural/tie--graph-guided-spatial-programs-need-uq.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--condition-module-can-corrupt-ranking-evaluation-indirectly.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--condition-module-can-corrupt-ranking-evaluation-indirectly.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--condition-module-can-corrupt-ranking-evaluation-indirectly.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--condition-module-can-corrupt-ranking-evaluation-indirectly.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--order-split-diagnosis-with-test-only-entities.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--order-split-diagnosis-with-test-only-entities.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--order-split-diagnosis-with-test-only-entities.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/episodic/anomalous--order-split-diagnosis-with-test-only-entities.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--distinguish-auroc-auprc-from-ranking-metrics-in-combination-prediction.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--distinguish-auroc-auprc-from-ranking-metrics-in-combination-prediction.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--distinguish-auroc-auprc-from-ranking-metrics-in-combination-prediction.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--distinguish-auroc-auprc-from-ranking-metrics-in-combination-prediction.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--do-not-overclaim-biological-priors-when-features-are-random-or-simplified.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--do-not-overclaim-biological-priors-when-features-are-random-or-simplified.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--do-not-overclaim-biological-priors-when-features-are-random-or-simplified.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--do-not-overclaim-biological-priors-when-features-are-random-or-simplified.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--open-order-generalization-as-order-composition.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--open-order-generalization-as-order-composition.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--open-order-generalization-as-order-composition.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--open-order-generalization-as-order-composition.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--picloss-edge-source-must-follow-split-regime.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--picloss-edge-source-must-follow-split-regime.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--picloss-edge-source-must-follow-split-regime.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/constraint-failure--picloss-edge-source-must-follow-split-regime.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--direct-composed-imitation-loss-for-order-experts.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--direct-composed-imitation-loss-for-order-experts.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--direct-composed-imitation-loss-for-order-experts.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--direct-composed-imitation-loss-for-order-experts.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--leave-one-out-contextualization-for-combination-entity-conditioning.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--leave-one-out-contextualization-for-combination-entity-conditioning.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--leave-one-out-contextualization-for-combination-entity-conditioning.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/operator-fail--leave-one-out-contextualization-for-combination-entity-conditioning.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--lexicographic-order-decomposition-for-unseen-k.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--lexicographic-order-decomposition-for-unseen-k.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--lexicographic-order-decomposition-for-unseen-k.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--lexicographic-order-decomposition-for-unseen-k.md
diff --git a/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--single-positive-versus-multi-positive-mrr-choice.md b/.agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--single-positive-versus-multi-positive-mrr-choice.md
similarity index 100%
rename from skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--single-positive-versus-multi-positive-mrr-choice.md
rename to .agents/skills/quantitative-biology/other-quantitative-biology/hzzzzzhappy/procedural/tie--single-positive-versus-multi-positive-mrr-choice.md
diff --git a/skills/statistics/computation/jamesyu420/procedural/operator-fail--gp-kernel-psd-and-jitter-check.md b/.agents/skills/statistics/computation/jamesyu420/procedural/operator-fail--gp-kernel-psd-and-jitter-check.md
similarity index 100%
rename from skills/statistics/computation/jamesyu420/procedural/operator-fail--gp-kernel-psd-and-jitter-check.md
rename to .agents/skills/statistics/computation/jamesyu420/procedural/operator-fail--gp-kernel-psd-and-jitter-check.md
diff --git a/skills/statistics/machine-learning/ergan-shang/procedural/constraint-failure--cross-validation-by-column-masking-in-rasch.md b/.agents/skills/statistics/machine-learning/ergan-shang/procedural/constraint-failure--cross-validation-by-column-masking-in-rasch.md
similarity index 100%
rename from skills/statistics/machine-learning/ergan-shang/procedural/constraint-failure--cross-validation-by-column-masking-in-rasch.md
rename to .agents/skills/statistics/machine-learning/ergan-shang/procedural/constraint-failure--cross-validation-by-column-masking-in-rasch.md
diff --git a/skills/statistics/machine-learning/ergan-shang/semantic/correction--bernoulli-logprob-loss-interpretation-in-rasch.md b/.agents/skills/statistics/machine-learning/ergan-shang/semantic/correction--bernoulli-logprob-loss-interpretation-in-rasch.md
similarity index 100%
rename from skills/statistics/machine-learning/ergan-shang/semantic/correction--bernoulli-logprob-loss-interpretation-in-rasch.md
rename to .agents/skills/statistics/machine-learning/ergan-shang/semantic/correction--bernoulli-logprob-loss-interpretation-in-rasch.md
diff --git a/skills/statistics/methodology/ergan-shang/procedural/constraint-failure--auc-computation-with-masked-matrix-data.md b/.agents/skills/statistics/methodology/ergan-shang/procedural/constraint-failure--auc-computation-with-masked-matrix-data.md
similarity index 100%
rename from skills/statistics/methodology/ergan-shang/procedural/constraint-failure--auc-computation-with-masked-matrix-data.md
rename to .agents/skills/statistics/methodology/ergan-shang/procedural/constraint-failure--auc-computation-with-masked-matrix-data.md
diff --git a/skills/statistics/methodology/jamesyu420/episodic/adaptation--mil-to-sgmm-when-spatial-clustering-matters.md b/.agents/skills/statistics/methodology/jamesyu420/episodic/adaptation--mil-to-sgmm-when-spatial-clustering-matters.md
similarity index 100%
rename from skills/statistics/methodology/jamesyu420/episodic/adaptation--mil-to-sgmm-when-spatial-clustering-matters.md
rename to .agents/skills/statistics/methodology/jamesyu420/episodic/adaptation--mil-to-sgmm-when-spatial-clustering-matters.md
diff --git a/skills/statistics/methodology/jamesyu420/semantic/non-public--psmil-bmle-smle-partial-labeling.md b/.agents/skills/statistics/methodology/jamesyu420/semantic/non-public--psmil-bmle-smle-partial-labeling.md
similarity index 100%
rename from skills/statistics/methodology/jamesyu420/semantic/non-public--psmil-bmle-smle-partial-labeling.md
rename to .agents/skills/statistics/methodology/jamesyu420/semantic/non-public--psmil-bmle-smle-partial-labeling.md
diff --git a/skills/statistics/methodology/jamesyu420/semantic/non-public--sgmm-spatially-varying-mixture-proportions.md b/.agents/skills/statistics/methodology/jamesyu420/semantic/non-public--sgmm-spatially-varying-mixture-proportions.md
similarity index 100%
rename from skills/statistics/methodology/jamesyu420/semantic/non-public--sgmm-spatially-varying-mixture-proportions.md
rename to .agents/skills/statistics/methodology/jamesyu420/semantic/non-public--sgmm-spatially-varying-mixture-proportions.md
diff --git a/AGENT.md b/AGENT.md
new file mode 100644
index 0000000..cdae280
--- /dev/null
+++ b/AGENT.md
@@ -0,0 +1,13 @@
+# ResearchSkills Agent Notes
+
+## Repository Layout
+
+- Public research skills live under `.agents/skills/`.
+- The `researchskills-extract/` package is the meta-skill tooling for extracting or converting skills; do not delete it when moving the public skill library.
+- Repository validators and contribution examples should point at `.agents/skills/`, not a top-level `skills/` directory.
+
+## Editing Rules
+
+- Keep README install instructions simple: use `npx skills add ScienceIntelligence/ResearchSkills` or `bunx skills add ScienceIntelligence/ResearchSkills`.
+- Preserve `/researchskills-extract` and `/researchskills-convert` documentation; those commands are for generating and submitting skills.
+- Use minimal diffs and avoid rewording unrelated project text.
diff --git a/readme.md b/README.md
similarity index 86%
rename from readme.md
rename to README.md
index 1577998..777b240 100644
--- a/readme.md
+++ b/README.md
@@ -54,7 +54,23 @@ Each skill encodes the knowledge, tools, reasoning protocols, and common pitfall
 
 ---
 
-<h2 align="center">2. How to Contribute</h2>
+<h2 align="center">2. Install ResearchSkills</h2>
+
+Install the public ResearchSkills library into your local agent:
+
+```bash
+npx skills add ScienceIntelligence/ResearchSkills
+```
+
+Or with Bun:
+
+```bash
+bunx skills add ScienceIntelligence/ResearchSkills
+```
+
+---
+
+<h2 align="center">3. How to Contribute</h2>
 
 ### Method A: Auto-Extract with `/researchskills-extract` (Recommended)
 
@@ -114,11 +130,45 @@ The command asks where your skills are, reads them, converts each one into the c
 
 </details>
 
+<details>
+<summary><strong>Method E: One-Click Distill Prompt for Claude Code / Codex</strong></summary>
+
+Use this when you already have a local skill, command, memory file, rubric, or project-specific instruction and want an agent to convert it into ResearchSkills format and submit it as a GitHub PR.
+
+Paste this into Claude Code or Codex from the repository root:
+
+```text
+Convert my existing skill material into ResearchSkills format and submit it via GitHub PR.
+
+Source material:
+- Ask me for the path(s) to my existing skill, command, memory, rubric, prompt, or instruction file.
+- Read only those files and any directly referenced local files that are necessary to understand them.
+
+Conversion rules:
+- Convert each reusable research know-how item into one ResearchSkills skill.
+- Use exactly one memory type per skill: procedural, semantic, or episodic.
+- Use exactly one subtype:
+  - procedural: tie, no-change, constraint-failure, operator-fail
+  - semantic: frontier, non-public, correction
+  - episodic: failure, adaptation, anomalous
+- Put files under `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`.
+- Preserve scientific content, but remove private paths, usernames, project names, private URLs, and collaborator names.
+- Skip generic engineering, DevOps, UI, database, Docker, git/npm, and textbook content unless it is directly research-method knowledge.
+
+PR workflow:
+- Create a branch.
+- Add only the converted skill files.
+- Run the repo validator or the closest available path check.
+- Open a GitHub PR with a short summary and the list of generated skills.
+```
+
+</details>
+
 > Don't see your field? [Propose a new area →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md) · Need a skill but can't write it yourself? [Request a skill →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=02-skill-request.yml)
 
 ---
 
-<h2 align="center">3. Skill Architecture</h2>
+<h2 align="center">4. Skill Architecture</h2>
 
 ResearchSkills skills are grounded in cognitive architecture theory — [Soar](https://en.wikipedia.org/wiki/Soar_(cognitive_architecture)) (Laird, 2012), [ACT-R](https://en.wikipedia.org/wiki/ACT-R) (Anderson, 1996), and [Case-Based Reasoning](https://en.wikipedia.org/wiki/Case-based_reasoning) (Kolodner, 1993). Skills are organized by **how researchers' minds actually store and retrieve expertise**, not by arbitrary categories.
 
@@ -166,7 +216,7 @@ Classified using Case-Based Reasoning terminology:
 ### Directory Structure
 
 ```
-skills/
+.agents/skills/
 └── {domain}/                    # 8 arXiv-aligned domains + management
     └── {subdomain}/             # 181 subcategories
         └── {contributor}/       # Your name
@@ -181,7 +231,7 @@ For the full rationale — why research is hard, why LLMs struggle with it, and
 
 ---
 
-<h2 align="center">4. Become a Reviewer</h2>
+<h2 align="center">5. Become a Reviewer</h2>
 
 Reviewers are domain experts who guard the scientific quality of skills in their subdomain. You need substantial peer-review experience in the relevant field.
 
@@ -193,7 +243,7 @@ Reviewers are domain experts who guard the scientific quality of skills in their
 
 ---
 
-<h2 align="center">5. Domains</h2>
+<h2 align="center">6. Domains</h2>
 
 <div align="center">
 
diff --git a/readme_zh.md b/readme_zh.md
index b138da5..6c8070a 100644
--- a/readme_zh.md
+++ b/readme_zh.md
@@ -1,6 +1,6 @@
 <div align="right">
 
-[English](readme.md) · [中文](#-researchskills)
+[English](README.md) · [中文](#-researchskills)
 
 </div>
 
@@ -52,7 +52,23 @@
 
 ---
 
-<h2 align="center">2. 如何贡献</h2>
+<h2 align="center">2. 安装 ResearchSkills</h2>
+
+把公开的 ResearchSkills 技能库安装到你的本地 agent：
+
+```bash
+npx skills add ScienceIntelligence/ResearchSkills
+```
+
+或使用 Bun：
+
+```bash
+bunx skills add ScienceIntelligence/ResearchSkills
+```
+
+---
+
+<h2 align="center">3. 如何贡献</h2>
 
 ### 方式 A：用 `/researchskills-extract` 自动提取（推荐）
 
@@ -183,11 +199,68 @@ fp16 精度下 1e-8 会被规约为 0，Adam 的更新步骤相当于除以零
 
 参照[这个指南 →**](https://researchskills.ai/submit-manually)
 
+<details>
+<summary><strong>方式 D：用 Claude Code / Codex 转换已有 Skills</strong></summary>
+
+已经有科研技能笔记、文档或任意格式的材料？运行一个命令，它会读取文件、转换为 ResearchSkills 格式，并打开 GitHub PR。
+
+```bash
+npm install -g @scienceintelligence/researchskills-extract
+```
+
+**Claude Code:**
+```
+/researchskills-convert
+```
+
+**Codex:**
+```
+$researchskills-convert
+```
+
+该命令会询问你的技能材料位置，逐条转换成规范格式，并自动处理 fork、分支、文件放置和脱敏。
+
+</details>
+
+<details>
+<summary><strong>方式 E：一键粘贴 Prompt 到 Claude Code / Codex 蒸馏已有技能</strong></summary>
+
+当你已经有本地 skill、command、memory、rubric、prompt 或项目指令，希望让 agent 把它转成 ResearchSkills 格式并通过 GitHub PR 提交时，使用这个方式。
+
+在仓库根目录，把下面这段粘贴到 Claude Code 或 Codex：
+
+```text
+请把我已有的技能材料转换为 ResearchSkills 格式，并通过 GitHub PR 提交。
+
+源材料：
+- 先问我要已有 skill、command、memory、rubric、prompt 或 instruction 文件路径。
+- 只读取这些文件，以及理解它们所必需的直接引用本地文件。
+
+转换规则：
+- 每个可复用的科研 know-how 转成一条 ResearchSkills skill。
+- 每条 skill 只能选择一种 memory type：procedural、semantic 或 episodic。
+- 每条 skill 只能选择一种 subtype：
+  - procedural：tie、no-change、constraint-failure、operator-fail
+  - semantic：frontier、non-public、correction
+  - episodic：failure、adaptation、anomalous
+- 文件放到 `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`。
+- 保留科学内容，但移除私有路径、用户名、项目名、私有 URL、合作者姓名。
+- 跳过通用工程、DevOps、UI、数据库、Docker、git/npm 和教科书内容，除非它直接属于科研方法知识。
+
+PR 流程：
+- 创建分支。
+- 只添加转换后的 skill 文件。
+- 运行仓库 validator 或最接近的路径检查。
+- 打开 GitHub PR，写清楚摘要和生成的 skill 列表。
+```
+
+</details>
+
 > 没有你的研究方向？[提议新领域 →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md) · 需要某个 Skill？[请求 Skill →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=02-skill-request.yml)
 
 ---
 
-<h2 align="center">3. Skill 架构设计</h2>
+<h2 align="center">4. Skill 架构设计</h2>
 
 ResearchSkills 的 Skill 设计基于认知架构理论 — [Soar](https://en.wikipedia.org/wiki/Soar_(cognitive_architecture)) (Laird, 2012)、[ACT-R](https://en.wikipedia.org/wiki/ACT-R) (Anderson, 1996) 和[基于案例的推理](https://en.wikipedia.org/wiki/Case-based_reasoning) (Kolodner, 1993)。Skill 按照**研究者大脑实际存储和检索专业知识的方式**来组织，而非任意分类。
 
@@ -235,7 +308,7 @@ ResearchSkills 的 Skill 设计基于认知架构理论 — [Soar](https://en.wi
 ### 目录结构
 
 ```
-skills/
+.agents/skills/
 └── {domain}/                    # 8 个 arXiv 对齐的顶层领域 + management
     └── {subdomain}/             # 181 个子领域
         └── {contributor}/       # 你的名字
diff --git a/researchskills-extract/commands/CONVERT-SKILL.md b/researchskills-extract/commands/CONVERT-SKILL.md
index ad70c4b..26e115e 100644
--- a/researchskills-extract/commands/CONVERT-SKILL.md
+++ b/researchskills-extract/commands/CONVERT-SKILL.md
@@ -178,7 +178,7 @@ git checkout -b "$BRANCH"
 Each skill must be placed at the correct path. **Create directories that don't exist.**
 
 ```
-skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
 ```
 
 Rules:
@@ -190,15 +190,15 @@ Rules:
 
 Example paths:
 ```
-skills/statistics/methodology/jdoe/procedural/constraint-failure--auc-computation-with-masked-data.md
-skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement-misconception.md
-skills/physics/geophysics/jdoe/episodic/adaptation--gradient-explosion-under-fp16.md
+.agents/skills/statistics/methodology/jdoe/procedural/constraint-failure--auc-computation-with-masked-data.md
+.agents/skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement-misconception.md
+.agents/skills/physics/geophysics/jdoe/episodic/adaptation--gradient-explosion-under-fp16.md
 ```
 
 For each skill:
 
 ```bash
-SKILL_DIR="skills/<domain>/<subdomain>/<contributor>/<memory_type>"
+SKILL_DIR=".agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>"
 mkdir -p "$SKILL_DIR"
 ```
 
@@ -207,7 +207,7 @@ Then write the skill file to `$SKILL_DIR/<subtype>--<skill-name>.md`.
 ### 5.5 — Commit
 
 ```bash
-git add skills/
+git add .agents/skills/
 git commit -m "add(skills): N skills by <contributor>"
 ```
 
@@ -232,7 +232,7 @@ gh pr create \
 <!-- Fill this in: one line per skill, e.g. "- procedural / constraint-failure — statistics/methodology — AUC Computation With Masked Data" -->
 
 ## Checklist
-- [x] Files placed in `skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
+- [x] Files placed in `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
 - [x] Filenames follow pattern `<subtype>--<skill-name>.md`
 - [x] All required frontmatter fields filled
 - [x] All required body sections present
diff --git a/researchskills-extract/commands/researchskills-convert.md b/researchskills-extract/commands/researchskills-convert.md
index af21a84..298e483 100644
--- a/researchskills-extract/commands/researchskills-convert.md
+++ b/researchskills-extract/commands/researchskills-convert.md
@@ -174,7 +174,7 @@ git checkout -b "$BRANCH"
 Each skill must be placed at the correct path. **Create directories that don't exist.**
 
 ```
-skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
 ```
 
 Rules:
@@ -186,15 +186,15 @@ Rules:
 
 Example paths:
 ```
-skills/statistics/methodology/jdoe/procedural/constraint-failure--auc-computation-with-masked-data.md
-skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement-misconception.md
-skills/physics/geophysics/jdoe/episodic/adaptation--gradient-explosion-under-fp16.md
+.agents/skills/statistics/methodology/jdoe/procedural/constraint-failure--auc-computation-with-masked-data.md
+.agents/skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement-misconception.md
+.agents/skills/physics/geophysics/jdoe/episodic/adaptation--gradient-explosion-under-fp16.md
 ```
 
 For each skill:
 
 ```bash
-SKILL_DIR="skills/<domain>/<subdomain>/<contributor>/<memory_type>"
+SKILL_DIR=".agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>"
 mkdir -p "$SKILL_DIR"
 ```
 
@@ -203,7 +203,7 @@ Then write the skill file to `$SKILL_DIR/<subtype>--<skill-name>.md`.
 ### 5.5 — Commit
 
 ```bash
-git add skills/
+git add .agents/skills/
 git commit -m "add(skills): N skills by <contributor>"
 ```
 
@@ -228,7 +228,7 @@ gh pr create \
 <!-- Fill this in: one line per skill, e.g. "- procedural / constraint-failure — statistics/methodology — AUC Computation With Masked Data" -->
 
 ## Checklist
-- [x] Files placed in `skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
+- [x] Files placed in `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
 - [x] Filenames follow pattern `<subtype>--<skill-name>.md`
 - [x] All required frontmatter fields filled
 - [x] All required body sections present
diff --git a/researchskills-extract/docs/codex-migration-plan.md b/researchskills-extract/docs/codex-migration-plan.md
index 0ef7408..d54f418 100644
--- a/researchskills-extract/docs/codex-migration-plan.md
+++ b/researchskills-extract/docs/codex-migration-plan.md
@@ -397,4 +397,4 @@ Output:
 ### Phase 3: 文档
 - [ ] 更新 `commands/researchskills-extract.md` 支持动态路径
 - [ ] 更新 `README.md` 双平台说明
-- [ ] 更新 `ResearchSkills/readme.md`
+- [ ] 更新 `ResearchSkills/README.md`
diff --git a/researchskills-extract/templates/skill-template.md b/researchskills-extract/templates/skill-template.md
index 9b07905..1df7925 100644
--- a/researchskills-extract/templates/skill-template.md
+++ b/researchskills-extract/templates/skill-template.md
@@ -16,7 +16,7 @@ status: draft                                # leave as draft; reviewer will upd
   1. Delete these comment blocks before submitting.
   2. Fill in ALL required frontmatter fields above.
   3. Complete each section below. "Required" sections must be present.
-  4. Run `python utils/tools/validate.py skills/<domain>/<subdomain>/<your-skill>.md` before opening a PR.
+  4. Run `python utils/tools/validate.py .agents/skills/<domain>/<subdomain>/<your-skill>.md` before opening a PR.
   5. See utils/SKILL_SCHEMA.md for full field documentation.
 
   NOTE: The primary contribution method is now decision trees via /researchskills-extract.
diff --git a/utils/CONTRIBUTING.md b/utils/CONTRIBUTING.md
index f7678f3..1368e14 100644
--- a/utils/CONTRIBUTING.md
+++ b/utils/CONTRIBUTING.md
@@ -57,21 +57,21 @@ Pick the template matching your skill type and copy it to the correct location:
 
 ```bash
 # Procedural — IF-THEN rules for research decisions
-cp skills/_template-procedural.md skills/<domain>/<subdomain>/<your-username>/procedural/<subtype>--<skill-name>.md
+cp .agents/skills/_template-procedural.md .agents/skills/<domain>/<subdomain>/<your-username>/procedural/<subtype>--<skill-name>.md
 
 # Semantic — domain facts LLMs don't reliably know
-cp skills/_template-semantic.md skills/<domain>/<subdomain>/<your-username>/semantic/<subtype>--<skill-name>.md
+cp .agents/skills/_template-semantic.md .agents/skills/<domain>/<subdomain>/<your-username>/semantic/<subtype>--<skill-name>.md
 
 # Episodic — concrete research episodes
-cp skills/_template-episodic.md skills/<domain>/<subdomain>/<your-username>/episodic/<subtype>--<skill-name>.md
+cp .agents/skills/_template-episodic.md .agents/skills/<domain>/<subdomain>/<your-username>/episodic/<subtype>--<skill-name>.md
 ```
 
 **File naming:** `<subtype>--<skill-name>.md`, lowercase, hyphen-separated. Examples:
-- `skills/physics/geophysics/jdoe/procedural/tie--exploiting-cross-domain-concepts.md`
-- `skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement.md`
-- `skills/physics/geophysics/jdoe/episodic/failure--gradient-explosion-under-fp16.md`
+- `.agents/skills/physics/geophysics/jdoe/procedural/tie--exploiting-cross-domain-concepts.md`
+- `.agents/skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement.md`
+- `.agents/skills/physics/geophysics/jdoe/episodic/failure--gradient-explosion-under-fp16.md`
 
-> The legacy template (`skills/_template.md`) is still supported for backward compatibility.
+> The legacy template (`.agents/skills/_template.md`) is still supported for backward compatibility.
 
 ### 3.3 Fill in the template
 
@@ -90,7 +90,7 @@ A domain reviewer listed in [CODEOWNERS](../.github/CODEOWNERS) will review your
 
 ## 3. Propose a New Area
 
-All 155 arXiv-aligned subcategory folders are pre-created under `skills/`. If you believe a subdomain is missing or want to propose a new top-level domain, [**open an issue →**](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md)
+All 155 arXiv-aligned subcategory folders are pre-created under `.agents/skills/`. If you believe a subdomain is missing or want to propose a new top-level domain, [**open an issue →**](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md)
 
 ---
 
@@ -179,18 +179,18 @@ cd ResearchSkills
 
 ```bash
 # 程序性 — 科研决策的 IF-THEN 规则
-cp skills/_template-procedural.md skills/<领域>/<子领域>/<用户名>/procedural/<子类型>--<技能名>.md
+cp .agents/skills/_template-procedural.md .agents/skills/<领域>/<子领域>/<用户名>/procedural/<子类型>--<技能名>.md
 
 # 语义性 — LLM 不可靠掌握的领域知识
-cp skills/_template-semantic.md skills/<领域>/<子领域>/<用户名>/semantic/<子类型>--<技能名>.md
+cp .agents/skills/_template-semantic.md .agents/skills/<领域>/<子领域>/<用户名>/semantic/<子类型>--<技能名>.md
 
 # 情景性 — 具体科研经历
-cp skills/_template-episodic.md skills/<领域>/<子领域>/<用户名>/episodic/<子类型>--<技能名>.md
+cp .agents/skills/_template-episodic.md .agents/skills/<领域>/<子领域>/<用户名>/episodic/<子类型>--<技能名>.md
 ```
 
 **命名规范：** `<子类型>--<技能名>.md`，小写字母，用连字符分隔。
 
-> 旧版模板（`skills/_template.md`）仍然支持向后兼容。
+> 旧版模板（`.agents/skills/_template.md`）仍然支持向后兼容。
 
 ### 3.3 填写模板
 
@@ -208,7 +208,7 @@ cp skills/_template-episodic.md skills/<领域>/<子领域>/<用户名>/episodic
 
 ## 4. 提议新领域或子领域
 
-所有 155 个 arXiv 对齐的子领域文件夹已预创建在 `skills/` 下。如果你认为缺少某个子领域或想提议新的顶层领域，[**提交 Issue →**](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md)
+所有 155 个 arXiv 对齐的子领域文件夹已预创建在 `.agents/skills/` 下。如果你认为缺少某个子领域或想提议新的顶层领域，[**提交 Issue →**](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md)
 
 ---
 
diff --git a/utils/SKILL_SCHEMA.md b/utils/SKILL_SCHEMA.md
index 218f15f..c3dbcf1 100644
--- a/utils/SKILL_SCHEMA.md
+++ b/utils/SKILL_SCHEMA.md
@@ -1,19 +1,19 @@
 # Skill Schema
 
-This document defines the specification for ResearchSkills skill files (manually written skills).
-
-> **Skills extracted via /researchskills-extract** are now the primary contribution format. See [Skill Schema Design](../docs/superpowers/specs/2026-04-11-skill-schema-design.md) for the current schema. This document covers only the manual skill file format.
+This document defines the current specification for ResearchSkills skill files.
 
 ---
 
 ## 1. File Location
 
 ```
-skills/<domain>/<subdomain>/<skill-name>.md
+.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
 ```
 
 - `domain` must match one of: `physics`, `mathematics`, `computer-science`, `quantitative-biology`, `statistics`, `eess`, `economics`, `quantitative-finance`, `management`
-- `subdomain` must match one of the arXiv-aligned subdomain folders under each domain (see `skills/<domain>/` for the full list)
+- `subdomain` must match one of the arXiv-aligned subdomain folders under each domain (see `.agents/skills/<domain>/` for the full list)
+- `contributor` should be the contributor's GitHub username or stable public handle
+- `memory_type` must be one of: `procedural`, `semantic`, `episodic`
 - `skill-name` must be lowercase, hyphen-separated
 
 ---
@@ -23,12 +23,11 @@ skills/<domain>/<subdomain>/<skill-name>.md
 ```yaml
 ---
 name: <string>                  # REQUIRED. Unique identifier. Lowercase, hyphen-separated.
-description: <string>           # REQUIRED. 1-2 sentences. When should this skill be invoked?
+memory_type: <string>           # REQUIRED. procedural | semantic | episodic
+subtype: <string>               # REQUIRED. Depends on memory_type.
 domain: <string>                # REQUIRED. One of: physics | mathematics | computer-science | quantitative-biology | statistics | eess | economics | quantitative-finance | management
-subdomain: <string>             # optional. More specific area within the domain.
-author: <string>                # REQUIRED. "Full Name (Affiliation)"
-expertise_level: <string>       # REQUIRED. One of: beginner | intermediate | advanced
-status: <string>                # REQUIRED. One of: draft | reviewed | verified
+subdomain: <string>             # REQUIRED. arXiv-aligned subdomain slug.
+contributor: <string>           # REQUIRED. GitHub username or stable public handle.
 ---
 ```
 
@@ -37,49 +36,76 @@ status: <string>                # REQUIRED. One of: draft | reviewed | verified
 | Field | Required | Type | Valid Values |
 |---|---|---|---|
 | `name` | yes | string | lowercase, hyphens only |
-| `description` | yes | string | 1-2 sentences |
+| `memory_type` | yes | enum | `procedural` `semantic` `episodic` |
+| `subtype` | yes | enum | see subtype table below |
 | `domain` | yes | enum | see list above |
-| `subdomain` | no | string | free text |
-| `author` | yes | string | "Name (Affiliation)" |
-| `expertise_level` | yes | enum | `beginner` `intermediate` `advanced` |
-| `status` | yes | enum | `draft` `reviewed` `verified` |
+| `subdomain` | yes | string | arXiv-aligned subdomain folder |
+| `contributor` | yes | string | GitHub username or public handle |
+
+### 2.2 Subtypes
+
+| Memory Type | Valid Subtypes |
+|---|---|
+| `procedural` | `tie`, `no-change`, `constraint-failure`, `operator-fail` |
+| `semantic` | `frontier`, `non-public`, `correction` |
+| `episodic` | `failure`, `adaptation`, `anomalous` |
 
 ---
 
 ## 3. Body Sections
 
-The skill body (after frontmatter) should follow this structure. Sections marked **Required** must be present.
+The skill body (after frontmatter) should follow the structure for its memory type.
 
-### 3.1 Required Sections
+### 3.1 Procedural
 
 ```markdown
-## Purpose
-One paragraph explaining what problem this skill solves and when to invoke it.
+## When
+Precise trigger conditions and exclusions.
 
-## Domain Knowledge
-Key concepts, definitions, equations, and established facts the AI needs to know.
+## Decision
+Preferred action, rejected alternatives, and reasoning.
 
-## Reasoning Protocol
-Numbered steps guiding the AI through the reasoning process for this domain.
+## Local Verifiers
+Concrete checks that the action is working.
 
-## Common Pitfalls
-Mistakes, misconceptions, and edge cases to avoid.
+## Failure Handling
+Fallback steps if the preferred action fails.
+
+## Anti-exemplars
+Situations where this skill should not be used.
 ```
 
-### 3.2 Recommended Sections
+### 3.2 Semantic
 
 ```markdown
-## Examples
-One or more worked examples demonstrating correct application of the skill.
+## Fact
+The specific fact, correction, or frontier knowledge.
+
+## Evidence
+Why this fact is known or credible.
+
+## LLM Default Belief
+For correction skills, what the model is likely to get wrong.
 
-## References
-Key papers, textbooks, or resources.
+## Expiry Signal
+When this fact may become stale.
 ```
 
----
+### 3.3 Episodic
+
+```markdown
+## Situation
+The concrete research situation.
 
-## 4. Extracted Skills Schema
+## Action
+What was tried or changed.
 
-The primary contribution format is now **research skills** — cognitive memory extracted from AI conversation history, organized into procedural (IF-THEN rules), semantic (domain facts), and episodic (research episodes) types.
+## Outcome
+What happened.
 
-For the full skill schema specification, see [docs/superpowers/specs/2026-04-11-skill-schema-design.md](../docs/superpowers/specs/2026-04-11-skill-schema-design.md).
+## Lesson
+The reusable lesson.
+
+## Retrieval Cues
+When an agent should recall this episode.
+```
diff --git a/utils/scripts/generate-taxonomy.js b/utils/scripts/generate-taxonomy.js
index a0927ca..4528f6a 100644
--- a/utils/scripts/generate-taxonomy.js
+++ b/utils/scripts/generate-taxonomy.js
@@ -2,7 +2,7 @@
 /**
  * generate-taxonomy.js
  *
- * Scan `skills/<domain>/<subdomain>/` directories and emit a JSON file
+ * Scan `.agents/skills/<domain>/<subdomain>/` directories and emit a JSON file
  * mapping domain → sorted list of subdomains. This is the single source of
  * truth for the arXiv-aligned taxonomy used by:
  *   - webserver review page (subdomain dropdown)
@@ -13,7 +13,7 @@
  *   node utils/scripts/generate-taxonomy.js
  *
  * Options:
- *   --skills <path>   directory to scan (default: ./skills)
+ *   --skills <path>   directory to scan (default: ./.agents/skills)
  *   --out <path>      JSON output path (default: ./webserver/public/taxonomy.json)
  *   --pretty          pretty-print JSON (default: true)
  */
@@ -24,7 +24,7 @@ const path = require('path');
 
 function parseArgs(argv) {
   const opts = {
-    skills: 'skills',
+    skills: path.join('.agents', 'skills'),
     out: path.join('webserver', 'public', 'taxonomy.json'),
     pretty: true,
   };
@@ -54,7 +54,7 @@ function main() {
   const outPath = path.resolve(opts.out);
 
   if (!isDir(skillsDir)) {
-    console.error(`error: skills directory not found: ${skillsDir}`);
+    console.error(`error: skill directory not found: ${skillsDir}`);
     process.exit(1);
   }
 
diff --git a/utils/tools/validate.py b/utils/tools/validate.py
index 36f9a89..b9fb211 100755
--- a/utils/tools/validate.py
+++ b/utils/tools/validate.py
@@ -2,6 +2,7 @@
 """Validate ResearchSkills skill files against the schema defined in SKILL_SCHEMA.md."""
 
 import sys
+import re
 from pathlib import Path
 
 try:
@@ -12,51 +13,66 @@
 
 REQUIRED_FIELDS = {
     "name": str,
-    "description": str,
+    "memory_type": str,
+    "subtype": str,
     "domain": str,
-    "author": str,
-    "expertise_level": str,
-    "status": str,
+    "subdomain": str,
+    "contributor": str,
 }
 
-VALID_EXPERTISE_LEVELS = {"beginner", "intermediate", "advanced"}
-VALID_STATUSES = {"draft", "reviewed", "verified"}
+VALID_MEMORY_TYPES = {"procedural", "semantic", "episodic"}
+VALID_SUBTYPES = {
+    "procedural": {"tie", "no-change", "constraint-failure", "operator-fail"},
+    "semantic": {"frontier", "non-public", "correction"},
+    "episodic": {"failure", "adaptation", "anomalous"},
+}
 
-REQUIRED_SECTIONS = [
-    "## Purpose",
-    "## Domain Knowledge",
-    "## Reasoning Protocol",
-    "## Common Pitfalls",
-]
+REQUIRED_SECTIONS = {
+    "semantic": ["## Fact"],
+    "episodic": ["## Situation", "## Action", "## Outcome"],
+}
 
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent
-SKILLS_DIR = REPO_ROOT / "skills"
+SKILLS_DIR = REPO_ROOT / ".agents" / "skills"
 
 # Derive valid domains from directory structure (single source of truth)
 VALID_DOMAINS = {p.name for p in SKILLS_DIR.iterdir() if p.is_dir() and not p.name.startswith('.')} if SKILLS_DIR.is_dir() else set()
 
 
+def slugify(value: str) -> str:
+    """Normalize display names to the filename slug convention."""
+    value = value.lower().replace("_", "-")
+    value = re.sub(r"[^a-z0-9]+", "-", value)
+    return value.strip("-")
+
+
 def validate_path(path: Path) -> list[str]:
-    """Validate that the skill file is in a correct skills/<domain>/<subdomain>/ directory."""
+    """Validate that the skill file is in a correct .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/ directory."""
     errors = []
     try:
         rel = path.resolve().relative_to(SKILLS_DIR)
     except ValueError:
-        errors.append(f"File is not under skills/ directory")
+        errors.append(f"File is not under .agents/skills/ directory")
         return errors
 
-    parts = rel.parts  # e.g. ('physics', 'quantum-physics', 'my-skill.md')
-    if len(parts) < 3:
-        errors.append(f"File must be in skills/<domain>/<subdomain>/, got: skills/{'/'.join(parts)}")
+    parts = rel.parts  # e.g. ('physics', 'geophysics', 'jdoe', 'semantic', 'correction--x.md')
+    if len(parts) < 5:
+        errors.append(f"File must be in .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/, got: .agents/skills/{'/'.join(parts)}")
         return errors
 
-    domain, subdomain = parts[0], parts[1]
+    domain, subdomain, contributor, memory_type = parts[0], parts[1], parts[2], parts[3]
     if domain not in VALID_DOMAINS:
         errors.append(f"Invalid domain folder '{domain}'. Must be one of: {sorted(VALID_DOMAINS)}")
 
     subdomain_dir = SKILLS_DIR / domain / subdomain
     if not subdomain_dir.is_dir():
-        errors.append(f"Subdomain folder 'skills/{domain}/{subdomain}/' does not exist")
+        errors.append(f"Subdomain folder '.agents/skills/{domain}/{subdomain}/' does not exist")
+
+    if not contributor:
+        errors.append("Contributor folder must not be empty")
+
+    if memory_type not in VALID_MEMORY_TYPES:
+        errors.append(f"Invalid memory type folder '{memory_type}'. Must be one of: {sorted(VALID_MEMORY_TYPES)}")
 
     return errors
 
@@ -100,20 +116,28 @@ def validate_file(path: Path) -> list[str]:
     if front.get("domain") and front["domain"] not in VALID_DOMAINS:
         errors.append(f"Invalid domain '{front['domain']}'. Must be one of: {sorted(VALID_DOMAINS)}")
 
-    if front.get("expertise_level") and front["expertise_level"] not in VALID_EXPERTISE_LEVELS:
-        errors.append(f"Invalid expertise_level '{front['expertise_level']}'. Must be one of: {sorted(VALID_EXPERTISE_LEVELS)}")
+    memory_type = front.get("memory_type")
+    subtype = front.get("subtype")
+    if memory_type and memory_type not in VALID_MEMORY_TYPES:
+        errors.append(f"Invalid memory_type '{memory_type}'. Must be one of: {sorted(VALID_MEMORY_TYPES)}")
 
-    if front.get("status") and front["status"] not in VALID_STATUSES:
-        errors.append(f"Invalid status '{front['status']}'. Must be one of: {sorted(VALID_STATUSES)}")
+    if memory_type in VALID_SUBTYPES and subtype and subtype not in VALID_SUBTYPES[memory_type]:
+        errors.append(f"Invalid subtype '{subtype}' for memory_type '{memory_type}'. Must be one of: {sorted(VALID_SUBTYPES[memory_type])}")
 
     # Check name matches filename
-    expected_name = path.stem
-    if front.get("name") and front["name"] != expected_name:
+    expected_name = path.stem.split("--", 1)[1] if "--" in path.stem else path.stem
+    if front.get("name") and slugify(front["name"]) != expected_name:
         errors.append(f"'name' field '{front['name']}' does not match filename '{expected_name}'")
 
     # Check required sections in body
     body = parts[2]
-    for section in REQUIRED_SECTIONS:
+    required_sections = REQUIRED_SECTIONS.get(memory_type, [])
+    if memory_type == "procedural":
+        has_decision_schema = ("## When" in body or "## When + Exclusions" in body) and "## Decision" in body
+        has_procedure_schema = "## Situation" in body and "## Procedure" in body
+        if not (has_decision_schema or has_procedure_schema):
+            errors.append("Missing procedural structure: use either '## When' + '## Decision' or '## Situation' + '## Procedure'")
+    for section in required_sections:
         if section not in body:
             errors.append(f"Missing required section: '{section}'")
 

From fdccb458ef35e1a617b7301315e48c79644632d5 Mon Sep 17 00:00:00 2001
From: LIghtJUNction <lightjunction.me@gmail.com>
Date: Mon, 1 Jun 2026 04:41:06 +0800
Subject: [PATCH 2/4] chore: replace extract package with meta skill

---
 .agents/skills/_template.md                   |   2 +-
 ...ls-meta-skill-extraction-and-conversion.md |  52 ++
 .github/CODEOWNERS                            |  46 +-
 .github/ISSUE_TEMPLATE/04-propose-new-area.md |   2 +-
 .github/pull_request_template.md              |   8 +-
 .github/workflows/onboard-maintainer.yml      |   8 +-
 AGENT.md                                      |   4 +-
 README.md                                     |  97 +-
 docs/decision-tree-v2-design.md               |   6 +-
 docs/why-research-is-hard.md                  |   4 +-
 readme_zh.md                                  |  97 +-
 researchskills-extract/README.md              | 114 ---
 .../commands/CONVERT-SKILL.md                 | 272 ------
 researchskills-extract/commands/SKILL.md      | 298 -------
 .../commands/researchskills-convert.md        | 268 ------
 .../commands/researchskills-extract.md        | 294 -------
 .../docs/codex-migration-plan.md              | 400 ---------
 ...26-04-13-review-and-score-skills-design.md | 299 -------
 researchskills-extract/package.json           |  41 -
 .../scripts/classify-projects.js              | 284 ------
 .../scripts/clean-skills.js                   | 202 -----
 .../scripts/extract-skills.js                 | 586 ------------
 researchskills-extract/scripts/finalize.js    | 175 ----
 .../scripts/format-session.js                 | 832 ------------------
 researchskills-extract/scripts/platform.js    | 195 ----
 researchskills-extract/scripts/postinstall.js | 204 -----
 .../scripts/postuninstall.js                  |  83 --
 .../scripts/scan-sessions.js                  | 598 -------------
 .../scripts/score-skills.js                   | 222 -----
 researchskills-extract/scripts/store-local.js | 231 -----
 .../scripts/upload-skills.js                  | 297 -------
 .../scripts/validate-skills.js                | 466 ----------
 researchskills-extract/templates/report.html  | 325 -------
 .../templates/skill-template.md               |  82 --
 .../tests/fixtures/engineering-session.jsonl  |   4 -
 .../tests/fixtures/minimal-session.jsonl      |   2 -
 .../tests/fixtures/research-session.jsonl     |   6 -
 .../tests/test-postinstall.js                 |  57 --
 utils/CONTRIBUTING.md                         |  48 +-
 utils/scripts/generate-taxonomy.js            |   2 +-
 40 files changed, 151 insertions(+), 7062 deletions(-)
 create mode 100644 .agents/skills/computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md
 delete mode 100644 researchskills-extract/README.md
 delete mode 100644 researchskills-extract/commands/CONVERT-SKILL.md
 delete mode 100644 researchskills-extract/commands/SKILL.md
 delete mode 100644 researchskills-extract/commands/researchskills-convert.md
 delete mode 100644 researchskills-extract/commands/researchskills-extract.md
 delete mode 100644 researchskills-extract/docs/codex-migration-plan.md
 delete mode 100644 researchskills-extract/docs/superpowers/specs/2026-04-13-review-and-score-skills-design.md
 delete mode 100644 researchskills-extract/package.json
 delete mode 100644 researchskills-extract/scripts/classify-projects.js
 delete mode 100644 researchskills-extract/scripts/clean-skills.js
 delete mode 100644 researchskills-extract/scripts/extract-skills.js
 delete mode 100644 researchskills-extract/scripts/finalize.js
 delete mode 100644 researchskills-extract/scripts/format-session.js
 delete mode 100644 researchskills-extract/scripts/platform.js
 delete mode 100644 researchskills-extract/scripts/postinstall.js
 delete mode 100644 researchskills-extract/scripts/postuninstall.js
 delete mode 100644 researchskills-extract/scripts/scan-sessions.js
 delete mode 100644 researchskills-extract/scripts/score-skills.js
 delete mode 100644 researchskills-extract/scripts/store-local.js
 delete mode 100755 researchskills-extract/scripts/upload-skills.js
 delete mode 100644 researchskills-extract/scripts/validate-skills.js
 delete mode 100644 researchskills-extract/templates/report.html
 delete mode 100644 researchskills-extract/templates/skill-template.md
 delete mode 100644 researchskills-extract/tests/fixtures/engineering-session.jsonl
 delete mode 100644 researchskills-extract/tests/fixtures/minimal-session.jsonl
 delete mode 100644 researchskills-extract/tests/fixtures/research-session.jsonl
 delete mode 100644 researchskills-extract/tests/test-postinstall.js

diff --git a/.agents/skills/_template.md b/.agents/skills/_template.md
index 2bc177e..17a9db1 100644
--- a/.agents/skills/_template.md
+++ b/.agents/skills/_template.md
@@ -19,7 +19,7 @@ status: draft                                # leave as draft; reviewer will upd
   4. Run `python utils/tools/validate.py .agents/skills/<domain>/<subdomain>/<your-skill>.md` before opening a PR.
   5. See utils/SKILL_SCHEMA.md for full field documentation.
 
-  NOTE: The primary contribution method is now skills via /researchskills-extract.
+  NOTE: The primary contribution method is now the ResearchSkills meta skill.
   This template is for manually written skills only.
 -->
 
diff --git a/.agents/skills/computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md b/.agents/skills/computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md
new file mode 100644
index 0000000..008e674
--- /dev/null
+++ b/.agents/skills/computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md
@@ -0,0 +1,52 @@
+---
+name: researchskills-meta-skill-extraction-and-conversion
+memory_type: procedural
+subtype: tie
+domain: computer-science
+subdomain: artificial-intelligence
+contributor: scienceintelligence
+---
+
+## When
+Use this meta skill when the task is to create ResearchSkills from existing material rather than to apply a domain skill directly. Typical inputs include AI conversation history, local agent skills, slash commands, memory files, rubrics, prompts, lab notes, research logs, or project instructions that may contain reusable research know-how.
+
+Do not use it for generic engineering, DevOps, UI, database, Docker, package-manager, git, or textbook material unless the content directly changes how research should be done.
+
+## Decision
+Treat extraction and conversion as a normal ResearchSkills skill workflow, not as a separate npm package or special command.
+
+Preferred:
+- Ask for the source path, conversation export, or pasted source material.
+- Read only the requested source and directly referenced files needed to understand it.
+- Extract one reusable research know-how item per output skill.
+- Assign exactly one memory type: `procedural`, `semantic`, or `episodic`.
+- Assign exactly one subtype:
+  - `procedural`: `tie`, `no-change`, `constraint-failure`, `operator-fail`
+  - `semantic`: `frontier`, `non-public`, `correction`
+  - `episodic`: `failure`, `adaptation`, `anomalous`
+- Place the generated file under `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`.
+- Remove private paths, usernames, private URLs, project names, and collaborator names while preserving scientific parameters, methods, model names, datasets, and mechanisms.
+- Validate the generated files, then submit them by GitHub PR.
+
+Rejected:
+- Do not install or document a separate npm package for this workflow.
+- Do not preserve legacy slash-command framing when a normal skill invocation is enough.
+- Do not merge multiple unrelated insights into one broad skill.
+- Do not create skills that only restate common LLM knowledge.
+
+Reasoning: the extraction workflow is itself a meta skill: it teaches an agent how to transform local knowledge into ResearchSkills. Keeping it in `.agents/skills/` makes installation, review, and contribution follow the same path as the rest of the library.
+
+## Local Verifiers
+- Every generated skill has frontmatter fields: `name`, `memory_type`, `subtype`, `domain`, `subdomain`, and `contributor`.
+- The file path matches its frontmatter memory type and subtype.
+- The body contains a concrete trigger, action or fact, reasoning mechanism, and failure or anti-example guidance when applicable.
+- Private identifiers are removed, but scientific content remains specific enough to be useful.
+- Repository validation passes for the generated skill files.
+
+## Failure Handling
+If the source material is too broad, ask for a narrower source directory, date range, topic, or conversation export before extracting. If the content is mostly generic engineering, report that no ResearchSkills-quality item was found and list the closest rejected candidates. If domain or subdomain is unclear, choose the closest arXiv-aligned folder and mention the uncertainty in the PR.
+
+## Anti-exemplars
+- A build failure fixed by changing an npm version is not a ResearchSkill unless it reveals a research-method constraint.
+- A prompt template for routine code review is not a ResearchSkill unless it encodes domain-specific research judgment.
+- A lab-specific secret, private endpoint, or personal workflow should be de-identified or skipped.
diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS
index a73f6a4..f7895b9 100644
--- a/.github/CODEOWNERS
+++ b/.github/CODEOWNERS
@@ -7,26 +7,26 @@
 *                              @HHHHHejia
 
 # Domain reviewers (arXiv-aligned) — fill in as experts are onboarded
-skills/physics/                @HHHHHejia   # TODO: replace with physics reviewer
-skills/physics/fluid-dynamics/  @qijiang-yoyo @HHHHHejia
-skills/physics/space-physics/  @huangzesen @HHHHHejia
-skills/physics/atmospheric-and-oceanic-physics/  @ktwu01 @HHHHHejia
-skills/physics/atmospheric-and-oceanic-physics/  @ktwu01 @HHHHHejia
-skills/physics/chemical-physics/  @Alexandrina-Chen @HHHHHejia
-skills/mathematics/            @HHHHHejia   # TODO: replace with math reviewer
-skills/computer-science/       @HHHHHejia   # TODO: replace with CS reviewer
-skills/computer-science/graphics/  @zjw49246 @HHHHHejia
-skills/computer-science/computation-and-language/  @hengzzzhou @HHHHHejia
-skills/computer-science/artificial-intelligence/  @zjw49246 @HHHHHejia
-skills/computer-science/hardware-architecture/  @Zhichenzzz @HHHHHejia
-skills/computer-science/multiagent-systems/  @wu-yc @HHHHHejia
-skills/computer-science/cryptography-and-security/  @xhyumiracle @HHHHHejia
-skills/quantitative-biology/   @HHHHHejia   # TODO: replace with q-bio reviewer
-skills/quantitative-biology/biomolecules/  @qijiang-yoyo @HHHHHejia
-skills/statistics/             @HHHHHejia   # TODO: replace with statistics reviewer
-skills/eess/                   @HHHHHejia   # TODO: replace with EESS reviewer
-skills/economics/              @HHHHHejia   # TODO: replace with economics reviewer
-skills/economics/general-economics/  @GalacEqua @HHHHHejia
-skills/quantitative-finance/   @HHHHHejia   # TODO: replace with q-fin reviewer
-skills/quantitative-finance/economics/  @Xinxin-m @HHHHHejia
-skills/management/             @HHHHHejia   # TODO: replace with management reviewer
+.agents/skills/physics/                @HHHHHejia   # TODO: replace with physics reviewer
+.agents/skills/physics/fluid-dynamics/  @qijiang-yoyo @HHHHHejia
+.agents/skills/physics/space-physics/  @huangzesen @HHHHHejia
+.agents/skills/physics/atmospheric-and-oceanic-physics/  @ktwu01 @HHHHHejia
+.agents/skills/physics/atmospheric-and-oceanic-physics/  @ktwu01 @HHHHHejia
+.agents/skills/physics/chemical-physics/  @Alexandrina-Chen @HHHHHejia
+.agents/skills/mathematics/            @HHHHHejia   # TODO: replace with math reviewer
+.agents/skills/computer-science/       @HHHHHejia   # TODO: replace with CS reviewer
+.agents/skills/computer-science/graphics/  @zjw49246 @HHHHHejia
+.agents/skills/computer-science/computation-and-language/  @hengzzzhou @HHHHHejia
+.agents/skills/computer-science/artificial-intelligence/  @zjw49246 @HHHHHejia
+.agents/skills/computer-science/hardware-architecture/  @Zhichenzzz @HHHHHejia
+.agents/skills/computer-science/multiagent-systems/  @wu-yc @HHHHHejia
+.agents/skills/computer-science/cryptography-and-security/  @xhyumiracle @HHHHHejia
+.agents/skills/quantitative-biology/   @HHHHHejia   # TODO: replace with q-bio reviewer
+.agents/skills/quantitative-biology/biomolecules/  @qijiang-yoyo @HHHHHejia
+.agents/skills/statistics/             @HHHHHejia   # TODO: replace with statistics reviewer
+.agents/skills/eess/                   @HHHHHejia   # TODO: replace with EESS reviewer
+.agents/skills/economics/              @HHHHHejia   # TODO: replace with economics reviewer
+.agents/skills/economics/general-economics/  @GalacEqua @HHHHHejia
+.agents/skills/quantitative-finance/   @HHHHHejia   # TODO: replace with q-fin reviewer
+.agents/skills/quantitative-finance/economics/  @Xinxin-m @HHHHHejia
+.agents/skills/management/             @HHHHHejia   # TODO: replace with management reviewer
diff --git a/.github/ISSUE_TEMPLATE/04-propose-new-area.md b/.github/ISSUE_TEMPLATE/04-propose-new-area.md
index 166ba25..9b4182f 100644
--- a/.github/ISSUE_TEMPLATE/04-propose-new-area.md
+++ b/.github/ISSUE_TEMPLATE/04-propose-new-area.md
@@ -12,7 +12,7 @@ assignees: HHHHHejia
 
 ## Type
 
-- [ ] New top-level domain (new folder directly under `skills/`)
+- [ ] New top-level domain (new folder directly under `.agents/skills/`)
 - [ ] New subdomain (subfolder within an existing domain — no approval needed, just include in your PR)
 
 ## Scope
diff --git a/.github/pull_request_template.md b/.github/pull_request_template.md
index 54149ae..a919434 100644
--- a/.github/pull_request_template.md
+++ b/.github/pull_request_template.md
@@ -1,7 +1,7 @@
 ## Submission Type
 
-- [ ] **Research Skill** — extracted via `/researchskills-extract`, converted via `/researchskills-convert`, or written by hand using the research skill templates
-- [ ] **Decision Tree** — research trajectory extracted via `/researchskills-extract` or web prompt
+- [ ] **Research Skill** — extracted or converted via the ResearchSkills meta skill, or written by hand using the research skill templates
+- [ ] **Decision Tree** — research trajectory extracted via the ResearchSkills meta skill or web prompt
 - [ ] **Skill (from Gemini/ChatGPT/Claude Web UI)** — submit via [here](https://researchskills.ai/submit-manually)
 - [ ] **Skill (legacy)** — manually written skill file using the legacy template (`_template.md`)
 
@@ -11,7 +11,7 @@
 
 **Domain / Subdomain:** <!-- e.g. physics / geophysics -->
 
-- [ ] Files are placed in `skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
+- [ ] Files are placed in `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
 - [ ] Filenames follow the pattern `<subtype>--<skill-name>.md` (lowercase, hyphen-separated)
 - [ ] All **required** frontmatter fields are filled in (`name`, `memory_type`, `subtype`, `domain`, `subdomain`, `contributor`)
 - [ ] All **required** body sections are present for each memory type:
@@ -34,7 +34,7 @@
 **Domain:** <!-- e.g. physics / quantum-physics -->
 **Skill name:** <!-- value of the `name` frontmatter field -->
 
-- [ ] File is placed in the correct `skills/<domain>/<subdomain>/` folder
+- [ ] File is placed in the correct `.agents/skills/<domain>/<subdomain>/` folder
 - [ ] Filename is lowercase and hyphen-separated
 - [ ] All **required** frontmatter fields are filled in (`name`, `description`, `domain`, `author`, `expertise_level`, `status: draft`)
 - [ ] All **required** body sections are present: Purpose, Domain Knowledge, Reasoning Protocol, Common Pitfalls
diff --git a/.github/workflows/onboard-maintainer.yml b/.github/workflows/onboard-maintainer.yml
index c7e4c68..090e05c 100644
--- a/.github/workflows/onboard-maintainer.yml
+++ b/.github/workflows/onboard-maintainer.yml
@@ -172,13 +172,13 @@ jobs:
           for row in $(echo "$SUBDOMAINS_JSON" | jq -r '.[] | @base64'); do
             DOMAIN_FOLDER=$(echo "$row" | base64 -d | jq -r '.domainFolder')
             SUBDOMAIN_FOLDER=$(echo "$row" | base64 -d | jq -r '.subdomainFolder')
-            NEW_LINE="skills/${DOMAIN_FOLDER}/${SUBDOMAIN_FOLDER}/  @${USERNAME} @HHHHHejia"
+            NEW_LINE=".agents/skills/${DOMAIN_FOLDER}/${SUBDOMAIN_FOLDER}/  @${USERNAME} @HHHHHejia"
 
-            if grep -q "^skills/${DOMAIN_FOLDER}/${SUBDOMAIN_FOLDER}/" "$CODEOWNERS_PATH"; then
+            if grep -q "^\\.agents/skills/${DOMAIN_FOLDER}/${SUBDOMAIN_FOLDER}/" "$CODEOWNERS_PATH"; then
               echo "CODEOWNERS entry for ${DOMAIN_FOLDER}/${SUBDOMAIN_FOLDER} already exists, updating..."
-              sed -i "s|^skills/${DOMAIN_FOLDER}/${SUBDOMAIN_FOLDER}/.*|${NEW_LINE}|" "$CODEOWNERS_PATH"
+              sed -i "s|^\\.agents/skills/${DOMAIN_FOLDER}/${SUBDOMAIN_FOLDER}/.*|${NEW_LINE}|" "$CODEOWNERS_PATH"
             else
-              DOMAIN_LINE=$(grep -n "^skills/${DOMAIN_FOLDER}/ " "$CODEOWNERS_PATH" | head -1 | cut -d: -f1)
+              DOMAIN_LINE=$(grep -n "^\\.agents/skills/${DOMAIN_FOLDER}/ " "$CODEOWNERS_PATH" | head -1 | cut -d: -f1)
               if [ -n "$DOMAIN_LINE" ]; then
                 sed -i "${DOMAIN_LINE}a\\${NEW_LINE}" "$CODEOWNERS_PATH"
               else
diff --git a/AGENT.md b/AGENT.md
index cdae280..01994a6 100644
--- a/AGENT.md
+++ b/AGENT.md
@@ -3,11 +3,11 @@
 ## Repository Layout
 
 - Public research skills live under `.agents/skills/`.
-- The `researchskills-extract/` package is the meta-skill tooling for extracting or converting skills; do not delete it when moving the public skill library.
+- The extraction/conversion workflow is a normal meta skill under `.agents/skills/`, not a separate package.
 - Repository validators and contribution examples should point at `.agents/skills/`, not a top-level `skills/` directory.
 
 ## Editing Rules
 
 - Keep README install instructions simple: use `npx skills add ScienceIntelligence/ResearchSkills` or `bunx skills add ScienceIntelligence/ResearchSkills`.
-- Preserve `/researchskills-extract` and `/researchskills-convert` documentation; those commands are for generating and submitting skills.
+- Describe extraction/conversion as using the ResearchSkills meta skill, not as legacy slash commands or separate packages.
 - Use minimal diffs and avoid rewording unrelated project text.
diff --git a/README.md b/README.md
index 777b240..0766770 100644
--- a/README.md
+++ b/README.md
@@ -32,7 +32,7 @@ This intuition lives in your head — the know-how, the heuristics, the reasonin
 
 **ResearchSkills captures it before it's lost.** We turn the tacit knowledge of the world's top researchers — their skills, thinking frameworks, and principles — into reusable AI agent skills (compatible with **Claude Code** and **Codex**). Every contribution makes every AI scientist — now and in the future — smarter, permanently.
 
-Each skill encodes the knowledge, tools, reasoning protocols, and common pitfalls of a scientific field. Skills can be written by domain experts or **auto-extracted from your research conversations** using `/researchskills-extract`. The command extracts three types of cognitive memory from your research sessions — **procedural** (IF-THEN rules for research impasses), **semantic** (facts LLMs don't know), and **episodic** (concrete research episodes) — then packages them as reusable skills. Point your AI agent at a skill, and it reasons like a domain expert.
+Each skill encodes the knowledge, tools, reasoning protocols, and common pitfalls of a scientific field. Skills can be written by domain experts or **auto-extracted from your research conversations** using the included ResearchSkills meta skill. The meta skill extracts three types of cognitive memory from your research sessions — **procedural** (IF-THEN rules for research impasses), **semantic** (facts LLMs don't know), and **episodic** (concrete research episodes) — then packages them as reusable skills. Point your AI agent at a skill, and it reasons like a domain expert.
 
 > **Note:** Applying a skill may trigger broad edits, long workflows, and significant token usage — review the expected scope before running one deeply.
 
@@ -72,97 +72,46 @@ bunx skills add ScienceIntelligence/ResearchSkills
 
 <h2 align="center">3. How to Contribute</h2>
 
-### Method A: Auto-Extract with `/researchskills-extract` (Recommended)
+### Method A: Use the ResearchSkills Meta Skill (Recommended)
 
-> **Prerequisite:** [Install Node.js](https://nodejs.org/) (includes npm). LTS version recommended.
+Install the library, then ask Claude Code, Codex, or another markdown-aware agent to use the `researchskills-meta-skill-extraction-and-conversion` skill. This is a **meta skill**: it turns conversation history, local skills, prompts, rubrics, notes, or other research know-how into ResearchSkills files.
 
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
+Example prompt:
 
-**Claude Code:**
-```
-/researchskills-extract
-```
+```text
+Use the ResearchSkills meta skill to extract or convert my research know-how into ResearchSkills files, validate them, and prepare a GitHub PR.
 
-**Codex** (start with `codex -a never -s danger-full-access`):
-```
-$researchskills-extract
-```
+Source material:
+- Ask me for the path(s), pasted text, or conversation export to use.
+- Read only that source and directly referenced files needed to understand it.
 
-> 💡 **For best results:** use the most powerful model with the highest reasoning effort — **Claude Code:** Opus 4.6 + max effort. **Codex:** GPT-5.4 + x-high. Don't worry about token usage — conversations are heavily compressed before analysis, and the per-session extraction is delegated to lighter models behind the scenes. Your chosen model mainly orchestrates the pipeline.
+Rules:
+- Create one ResearchSkills file per reusable research know-how item.
+- Use exactly one memory type: procedural, semantic, or episodic.
+- Use exactly one valid subtype for that memory type.
+- Put files under `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`.
+- Preserve scientific content, but remove private paths, usernames, project names, private URLs, and collaborator names.
+- Skip generic engineering, DevOps, UI, database, Docker, git/npm, and textbook content unless it is directly research-method knowledge.
+```
 
-The command scans your conversation history and extracts **research skills** organized by cognitive memory type:
+The meta skill extracts **research skills** organized by cognitive memory type:
 
 - **Procedural memory:** IF-THEN rules for navigating research impasses (e.g., "IF gradient explodes THEN check learning rate before architecture")
 - **Semantic memory:** Domain facts that LLMs don't reliably know (e.g., calibration constants, method limitations, undocumented tool behaviors)
 - **Episodic memory:** Concrete research episodes capturing what was tried, what failed, and what the researcher learned
 
-An interactive browser review page lets you verify the extracted skills, check de-identification, and bind them to your paper (arXiv/DOI) or project. Submit your skills to ResearchSkills, where they become part of a growing knowledge base for building better AI scientists.
+Review the generated files, check de-identification, validate them, and submit them by GitHub PR.
 
 ### Method B: One-Click Prompt for Web Users (ChatGPT / Claude / Gemini)
 After running, submit via [**here →**](https://researchskills.ai/submit-manually/#auto-parse)
 
-### Method C: Write Manually
-
-Write your own skill following the [**guide →**](https://researchskills.ai/submit-manually/#manual-entry)
-
-<details>
-<summary><strong>Method D: Convert Existing Skills via Claude Code / Codex</strong></summary>
-
-Already have research skills in notes, documents, or any format? Run one command — it reads your files, converts them, and opens a PR.
-
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
-
-**Claude Code:**
-```
-/researchskills-convert
-```
-
-**Codex:**
-```
-$researchskills-convert
-```
-
-The command asks where your skills are, reads them, converts each one into the correct format, and opens a PR to this repository. Forking, branching, file placement, and de-identification are handled automatically.
-
-</details>
+### Method C: Convert Existing Skills
 
-<details>
-<summary><strong>Method E: One-Click Distill Prompt for Claude Code / Codex</strong></summary>
+Already have a local skill, command, memory file, rubric, prompt, project instruction, note, or document? Use the same ResearchSkills meta skill from Method A. It converts existing material into `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md` files and prepares a PR.
 
-Use this when you already have a local skill, command, memory file, rubric, or project-specific instruction and want an agent to convert it into ResearchSkills format and submit it as a GitHub PR.
+### Method D: Write Manually
 
-Paste this into Claude Code or Codex from the repository root:
-
-```text
-Convert my existing skill material into ResearchSkills format and submit it via GitHub PR.
-
-Source material:
-- Ask me for the path(s) to my existing skill, command, memory, rubric, prompt, or instruction file.
-- Read only those files and any directly referenced local files that are necessary to understand them.
-
-Conversion rules:
-- Convert each reusable research know-how item into one ResearchSkills skill.
-- Use exactly one memory type per skill: procedural, semantic, or episodic.
-- Use exactly one subtype:
-  - procedural: tie, no-change, constraint-failure, operator-fail
-  - semantic: frontier, non-public, correction
-  - episodic: failure, adaptation, anomalous
-- Put files under `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`.
-- Preserve scientific content, but remove private paths, usernames, project names, private URLs, and collaborator names.
-- Skip generic engineering, DevOps, UI, database, Docker, git/npm, and textbook content unless it is directly research-method knowledge.
-
-PR workflow:
-- Create a branch.
-- Add only the converted skill files.
-- Run the repo validator or the closest available path check.
-- Open a GitHub PR with a short summary and the list of generated skills.
-```
-
-</details>
+Write your own skill following the [**guide →**](https://researchskills.ai/submit-manually/#manual-entry)
 
 > Don't see your field? [Propose a new area →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md) · Need a skill but can't write it yourself? [Request a skill →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=02-skill-request.yml)
 
diff --git a/docs/decision-tree-v2-design.md b/docs/decision-tree-v2-design.md
index cde54e9..d9eabd3 100644
--- a/docs/decision-tree-v2-design.md
+++ b/docs/decision-tree-v2-design.md
@@ -1,10 +1,10 @@
 > **Deprecated:** This document describes the Phase 1 decision tree format, which has been replaced by the cognitive memory-type skill system. See [Skill Schema Design](superpowers/specs/2026-04-11-skill-schema-design.md) for the current design.
 
-# /researchskills-extract v2 — Research Decision Tree Collection
+# ResearchSkills Meta Skill v2 — Research Decision Tree Collection
 
 ## Context
 
-v1 of `/researchskills-extract` treats chat history as uniform text and summarizes it into 10 fixed categories. This produces textbook-like output that misses the actual value: the tacit judgments researchers make during their work — what they tried, what they rejected, and why.
+v1 of the ResearchSkills extraction workflow treats chat history as uniform text and summarizes it into 10 fixed categories. This produces textbook-like output that misses the actual value: the tacit judgments researchers make during their work — what they tried, what they rejected, and why.
 
 **Strategic insight:** The scarcest resource is researcher participation, not algorithm precision. Rather than designing the perfect skill-extraction pipeline first, v2 prioritizes **collecting structured-but-flexible data at low friction**. We can iterate on downstream analysis later without re-collecting.
 
@@ -93,7 +93,7 @@ Actions interleave freely — the phase grouping is for human readability, not a
 
 ### Step 1: Local Extraction
 
-User runs `/researchskills-extract` in Claude Code or Codex. The system:
+User invokes the ResearchSkills meta skill in Claude Code or Codex. The system:
 
 1. Reads local conversation history.
 2. For each conversation, the LLM:
diff --git a/docs/why-research-is-hard.md b/docs/why-research-is-hard.md
index 82085b8..d1f5c2a 100644
--- a/docs/why-research-is-hard.md
+++ b/docs/why-research-is-hard.md
@@ -1,6 +1,6 @@
 # 为什么科研是困难的，以及我们需要什么样的 Research Skills
 
-这是一份 problem statement。它的具体目的是把一件事说清楚——ResearchSkills 的 `/researchskills-extract` v2 应当被设计成什么样——同时把这件事所依赖的几个前提，按七个连续问题摊开来。
+这是一份 problem statement。它的具体目的是把一件事说清楚——ResearchSkills 的 meta skill 提取流程应当被设计成什么样——同时把这件事所依赖的几个前提，按七个连续问题摊开来。
 
 1. [为什么科研本身如此困难](#q1-为什么科研是困难的)
 2. [为什么大模型并未因此自然变成科研代理](#q2-为什么现在的大模型自动化科研效果不佳)
@@ -334,7 +334,7 @@ skill 的设计应当直接对应科研的结构性困难。既然科研难在
 
 提取的目标是从这些材料背后反推一个稳定的中层策略，而非复述材料本身。我们想恢复的是一个在状态 $h_t$ 下能够影响下一步动作选择的 skill $\sigma$，而非某篇论文说了什么或某段对话逐字说了什么。这件事的难点是抽取 policy，文字本身的抽取相对容易。
 
-下面是 `/researchskills-extract` v2 的具体设计。它把上面这个目标拆成四个组件——A、C、B 和一个最终的聚类层——并由一个 Step 0 的纸驱动语义筛选作为入口。
+下面是 ResearchSkills meta skill 提取流程的具体设计。它把上面这个目标拆成四个组件——A、C、B 和一个最终的聚类层——并由一个 Step 0 的纸驱动语义筛选作为入口。
 
 ### Step 0：纸驱动的语义筛选
 
diff --git a/readme_zh.md b/readme_zh.md
index 6c8070a..9f5aab9 100644
--- a/readme_zh.md
+++ b/readme_zh.md
@@ -32,7 +32,7 @@
 
 **ResearchSkills 在它消失之前把它留住。** 我们把全世界顶尖研究者的隐性知识 — 他们的技能、思维框架和原则 — 变成可复用的 AI agent 技能（兼容 **Claude Code** 和 **Codex**）。每一份贡献，都让现在和未来的每一个 AI 科学家变得更聪明，永久地。
 
-每个 Skill 编码了领域知识、工具、推理协议和常见陷阱。Skill 可以由领域专家手动撰写，也可以通过 `/researchskills-extract` **从你的科研对话中自动提取**。该命令从你的科研会话中提取三种认知记忆 — **程序性记忆**（应对科研困境的 IF-THEN 规则）、**语义记忆**（LLM 不知道的领域事实）和**情景记忆**（具体的科研经历）— 并将它们打包为可复用的 Skill。让 AI 调用一个 Skill，就能像领域专家一样思考。
+每个 Skill 编码了领域知识、工具、推理协议和常见陷阱。Skill 可以由领域专家手动撰写，也可以通过内置的 ResearchSkills meta skill **从你的科研对话中自动提取**。这个 meta skill 会从你的科研会话中提取三种认知记忆 — **程序性记忆**（应对科研困境的 IF-THEN 规则）、**语义记忆**（LLM 不知道的领域事实）和**情景记忆**（具体的科研经历）— 并将它们打包为可复用的 Skill。让 AI 调用一个 Skill，就能像领域专家一样思考。
 
 ---
 
@@ -70,33 +70,35 @@ bunx skills add ScienceIntelligence/ResearchSkills
 
 <h2 align="center">3. 如何贡献</h2>
 
-### 方式 A：用 `/researchskills-extract` 自动提取（推荐）
+### 方式 A：使用 ResearchSkills Meta Skill（推荐）
 
-> **前置条件：** [安装 Node.js](https://nodejs.org/)（自带 npm）。推荐 LTS 版本。
+安装技能库后，让 Claude Code、Codex 或任何能读取 markdown skill 的 agent 使用 `researchskills-meta-skill-extraction-and-conversion`。这是一个 **meta skill**：它把对话历史、本地 skill、prompt、rubric、笔记或其他科研 know-how 转成 ResearchSkills 文件。
 
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
+示例 prompt：
 
-**Claude Code:**
-```
-/researchskills-extract
-```
+```text
+请使用 ResearchSkills meta skill，把我的科研 know-how 提取或转换成 ResearchSkills 文件，验证后准备 GitHub PR。
 
-**Codex:**
-```
-$researchskills-extract
-```
+源材料：
+- 先问我要使用的路径、粘贴文本或对话导出。
+- 只读取这些源材料，以及理解它们所必需的直接引用文件。
 
-> 💡 **为了最佳效果：** 使用最强模型 + 最高推理强度 — **Claude Code：** Opus 4.6 + max effort。**Codex：** GPT-5.4 + x-high。不必担心 token 消耗 — 对话会被充分压缩后再分析，每个会话的提取会交给较小的模型处理。你选的模型主要负责编排流水线。
+规则：
+- 每个可复用科研 know-how 生成一个 ResearchSkills 文件。
+- 每条只选择一种 memory type：procedural、semantic 或 episodic。
+- 每条只选择该 memory type 下的一个合法 subtype。
+- 文件放到 `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`。
+- 保留科学内容，但移除私有路径、用户名、项目名、私有 URL、合作者姓名。
+- 跳过通用工程、DevOps、UI、数据库、Docker、git/npm 和教科书内容，除非它直接属于科研方法知识。
+```
 
-该命令会扫描你的对话历史，提取按认知记忆类型组织的**科研技能**：
+这个 meta skill 会提取按认知记忆类型组织的**科研技能**：
 
 - **程序性记忆：** 应对科研困境的 IF-THEN 规则（如"IF 梯度爆炸 THEN 先检查学习率再改架构"）
 - **语义记忆：** LLM 不可靠掌握的领域事实（如校准常数、未记录的工具行为、方法局限性）
 - **情景记忆：** 具体的科研经历，记录尝试了什么、失败了什么、学到了什么
 
-浏览器交互页面让你审核提取的技能、检查脱敏处理、绑定论文（arXiv/DOI）或项目。提交你的技能到 ResearchSkills，它将成为构建更强 AI 科学家的知识库的一部分。
+审核生成的文件、检查脱敏处理、运行验证，然后通过 GitHub PR 提交。
 
 ### 方式 B：网页版用户一键提取（ChatGPT / Claude / Gemini）
 
@@ -195,66 +197,13 @@ fp16 精度下 1e-8 会被规约为 0，Adam 的更新步骤相当于除以零
 
 运行后提交：[**提交你的 Skill →**](https://researchskills.ai/)
 
-### 方式 C：手动撰写
-
-参照[这个指南 →**](https://researchskills.ai/submit-manually)
-
-<details>
-<summary><strong>方式 D：用 Claude Code / Codex 转换已有 Skills</strong></summary>
-
-已经有科研技能笔记、文档或任意格式的材料？运行一个命令，它会读取文件、转换为 ResearchSkills 格式，并打开 GitHub PR。
-
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
-
-**Claude Code:**
-```
-/researchskills-convert
-```
-
-**Codex:**
-```
-$researchskills-convert
-```
+### 方式 C：转换已有 Skills
 
-该命令会询问你的技能材料位置，逐条转换成规范格式，并自动处理 fork、分支、文件放置和脱敏。
+已经有本地 skill、command、memory、rubric、prompt、项目指令、笔记或文档？仍然使用方式 A 的 ResearchSkills meta skill。它会把已有材料转换成 `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md` 文件，并准备 PR。
 
-</details>
+### 方式 D：手动撰写
 
-<details>
-<summary><strong>方式 E：一键粘贴 Prompt 到 Claude Code / Codex 蒸馏已有技能</strong></summary>
-
-当你已经有本地 skill、command、memory、rubric、prompt 或项目指令，希望让 agent 把它转成 ResearchSkills 格式并通过 GitHub PR 提交时，使用这个方式。
-
-在仓库根目录，把下面这段粘贴到 Claude Code 或 Codex：
-
-```text
-请把我已有的技能材料转换为 ResearchSkills 格式，并通过 GitHub PR 提交。
-
-源材料：
-- 先问我要已有 skill、command、memory、rubric、prompt 或 instruction 文件路径。
-- 只读取这些文件，以及理解它们所必需的直接引用本地文件。
-
-转换规则：
-- 每个可复用的科研 know-how 转成一条 ResearchSkills skill。
-- 每条 skill 只能选择一种 memory type：procedural、semantic 或 episodic。
-- 每条 skill 只能选择一种 subtype：
-  - procedural：tie、no-change、constraint-failure、operator-fail
-  - semantic：frontier、non-public、correction
-  - episodic：failure、adaptation、anomalous
-- 文件放到 `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`。
-- 保留科学内容，但移除私有路径、用户名、项目名、私有 URL、合作者姓名。
-- 跳过通用工程、DevOps、UI、数据库、Docker、git/npm 和教科书内容，除非它直接属于科研方法知识。
-
-PR 流程：
-- 创建分支。
-- 只添加转换后的 skill 文件。
-- 运行仓库 validator 或最接近的路径检查。
-- 打开 GitHub PR，写清楚摘要和生成的 skill 列表。
-```
-
-</details>
+参照[这个指南 →**](https://researchskills.ai/submit-manually)
 
 > 没有你的研究方向？[提议新领域 →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=04-propose-new-area.md) · 需要某个 Skill？[请求 Skill →](https://github.com/ScienceIntelligence/ResearchSkills/issues/new?template=02-skill-request.yml)
 
diff --git a/researchskills-extract/README.md b/researchskills-extract/README.md
deleted file mode 100644
index c35bb48..0000000
--- a/researchskills-extract/README.md
+++ /dev/null
@@ -1,114 +0,0 @@
-# @scienceintelligence/researchskills-extract
-
-> Automatically extract **research skills** from Claude Code / Codex conversation history and submit them to [ResearchSkills](https://github.com/ScienceIntelligence/ResearchSkills).
-
-## What It Does
-
-When you use Claude Code or Codex for scientific research — data analysis, paper writing, experiment design, theoretical derivation — your conversations contain valuable tacit knowledge: judgment calls, abandoned approaches, tool choices, and reasoning patterns.
-
-`/researchskills-extract` extracts three types of cognitive memory from your research sessions:
-
-- **Procedural memory:** IF-THEN rules for navigating research impasses (e.g., "IF gradient explodes THEN check learning rate before architecture")
-- **Semantic memory:** Domain facts that LLMs don't reliably know (e.g., calibration constants, undocumented tool behaviors)
-- **Episodic memory:** Concrete research episodes capturing what was tried, what failed, and what was learned
-
-## Install
-
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
-
-This installs the command automatically to both platforms:
-- **Claude Code** → `~/.claude/commands/researchskills-extract.md`
-- **Codex** → `~/.codex/skills/researchskills-extract/SKILL.md`
-
-## Usage
-
-**Claude Code:**
-```
-/researchskills-extract
-```
-
-**Codex** (start with `codex -a never -s danger-full-access`):
-```
-$researchskills-extract
-```
-
-> 💡 **For best results:** use the most powerful model with the highest reasoning effort — **Claude Code:** Opus 4.6 + max effort. **Codex:** GPT-5.4 + x-high. Don't worry about token usage — conversations are heavily compressed before analysis, and per-session extraction is delegated to lighter models behind the scenes. Your chosen model mainly orchestrates the pipeline.
-
-The command runs a 7-stage pipeline:
-
-1. **Scan** — discover all Claude Code and Codex sessions
-2. **Classify** — identify research vs. engineering projects (Sonnet)
-3. **Confirm** — you choose which projects to scan (multi-select)
-4. **Extract** — extract research skills per session (Sonnet), organized by cognitive memory type
-5. **Clean** — review extracted skills with Opus: reject engineering content, fix PII, merge duplicates
-6. **Score** — assess each skill's value on 3 dimensions with Opus (procedural / semantic / episodic, 0-5)
-7. **Finalize** — upload cleaned, scored skills to [researchskills.ai](https://researchskills.ai)
-8. **Summary** — report results with review statistics
-
-## Output
-
-Each skill is a markdown file with YAML frontmatter, including three review scores:
-
-```yaml
----
-name: gradient-explosion-diagnosis
-memory_type: procedural
-subtype: operator-fail
-llm_score: 4
-review_scores:
-  procedural: 4   # decision frameworks AI doesn't know
-  semantic: 2      # facts/beliefs AI doesn't have
-  episodic: 3      # concrete research experiences
-tags: [gradient-descent, debugging, neural-networks]
-domain: computer-science
-subdomain: machine-learning
-contributor: anon-7f3b42c9
----
-
-## When
-Using Adam or SGD with deep networks; loss spikes unpredictably.
-
-## Decision
-Check learning rate first (most common cause), not architecture.
-
-## Local Verifiers
-- nan_count in gradients > threshold
-- loss jump > 10x in single step
-
-## Failure Handling
-If gradient norm clipping doesn't fix: check batch normalization placement
-```
-
-## Contributing Back
-
-After extraction, an interactive review page opens at `researchskills.ai/review/batch/<id>` where you can:
-
-- Review and edit skill content
-- See the 3-dimension review scores
-- Assign domain/subdomain taxonomy
-- Submit to ResearchSkills
-
-## Uninstall
-
-```bash
-npm uninstall -g @scienceintelligence/researchskills-extract
-```
-
-## Privacy
-
-- All analysis happens locally via your Claude Code / Codex session
-- Session data is read from `~/.claude/projects/` and `~/.codex/` on your machine
-- **You choose which projects to scan** — the tool pauses after classification for your selection
-- Unselected projects are skipped for extraction (classification reads only brief message samples)
-- AI auto-strips personal information; you review before submitting
-- Nothing is uploaded without your explicit consent
-
-## License
-
-[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)
-
-## Part of [ResearchSkills](https://github.com/ScienceIntelligence/ResearchSkills)
-
-> Building the Library of Alexandria for AGI — Accelerating Automated Scientific Discovery.
diff --git a/researchskills-extract/commands/CONVERT-SKILL.md b/researchskills-extract/commands/CONVERT-SKILL.md
deleted file mode 100644
index 26e115e..0000000
--- a/researchskills-extract/commands/CONVERT-SKILL.md
+++ /dev/null
@@ -1,272 +0,0 @@
----
-name: "researchskills-convert"
-description: "Convert existing research skills from any format (notes, documents, other AI tools) into ResearchSkills format, then fork the repo and open a PR. Use when the user says 'convert my skills', 'submit existing skills', 'I already have research notes', 'turn my notes into ResearchSkills', 'convert from another format', or wants to contribute skills they already wrote in a different format."
----
-# /researchskills-convert
-
-Convert existing research skills from any format (notes, documents, other AI tools) into **ResearchSkills** format, then fork the repo and open a PR — all in one step.
-
-**This is an interactive skill.** Ask the user ONE question at a time. Do not dump all questions at once.
-
----
-
-## Stage 1 — Collect Input
-
-Ask the user:
-
-> **Where are your skills?**
-> Provide a file path, a directory, or paste them here.
-
-Read the files or accept the pasted text. If a directory, recursively read all `.md`, `.txt`, `.json`, and `.yaml` files in it.
-
----
-
-## Stage 2 — Identify the User
-
-Get the GitHub username for the `contributor` field:
-
-```bash
-gh api user --jq '.login'
-```
-
-If `gh` is not installed or not authenticated, tell the user:
-
-> `gh` (GitHub CLI) is required for PR creation. Install it:
-> - macOS: `brew install gh`
-> - Linux: see https://cli.github.com
->
-> Then run `gh auth login` to authenticate.
-
-**Stop here until `gh api user` succeeds.**
-
----
-
-## Stage 3 — Convert Skills
-
-Analyze the input and convert each genuine **research** skill into the ResearchSkills format. Apply these filters:
-
-### What to INCLUDE
-- Domain-specific research insights a researcher would find non-obvious
-- Methodology decisions, scientific corrections, experimental lessons
-- Knowledge that is frontier, non-public, or corrects LLM misconceptions
-
-### What to SKIP
-- Generic software engineering (git workflows, CI/CD, Docker, deployment)
-- DevOps, infrastructure, or tooling knowledge
-- Textbook material readily available in LLM training data
-- Content that is not related to scientific research
-
-### Memory Types and Required Sections
-
-**Procedural** (subtypes: `tie`, `no-change`, `constraint-failure`, `operator-fail`)
-- When — trigger conditions and exclusions
-- Decision — Preferred, Rejected, Reasoning
-- Local Verifiers — concrete diagnostics
-- Failure Handling — fallback strategies
-- Anti-exemplars — when NOT to use (recommended)
-
-**Semantic** (subtypes: `frontier`, `non-public`, `correction`)
-- Fact — precise core claim
-- Evidence — how you know this is true
-- LLM Default Belief — **correction subtype only**, delete for others
-- Expiry Signal — when to revisit
-
-**Episodic** (subtypes: `failure`, `adaptation`, `anomalous`)
-- Situation — tools, dataset, parameters, expectations
-- Action — what was done, with specifics
-- Outcome — concrete result with metrics
-- Lesson — specific IF-THEN rule
-- Retrieval Cues — trigger conditions for recall
-
-### Frontmatter
-
-Every skill file must have this exact frontmatter:
-
-```yaml
----
-name: "Skill Name In Title Case"
-memory_type: procedural          # procedural | semantic | episodic
-subtype: tie                     # see subtypes above
-domain: computer-science         # see domain list below
-subdomain: machine-learning      # arXiv-aligned subdomain
-contributor: gh-username          # from Stage 2
----
-```
-
-### Valid Domains
-
-Pick from these arXiv-aligned domains:
-- `physics`
-- `mathematics`
-- `computer-science`
-- `quantitative-biology`
-- `statistics`
-- `eess` (electrical engineering and systems science)
-- `economics`
-- `quantitative-finance`
-
-For subdomain, use an arXiv-aligned subcategory (e.g., `machine-learning`, `geophysics`, `genomics`, `methodology`).
-
-### De-identification
-
-Before outputting, strip:
-- Personal names (replace with role descriptions)
-- Private file paths (use generic paths)
-- Internal URLs and hostnames
-- Lab-specific identifiers
-- API keys, tokens, credentials
-
-All output MUST be in English. If source material is in another language, paraphrase in English.
-
----
-
-## Stage 4 — Show Preview
-
-Show the user all converted skills in a summary table:
-
-```
-Converted N skills:
-
-  #  Type         Subtype              Domain / Subdomain                   Name
-  1  procedural   constraint-failure   statistics / methodology             AUC Computation With Masked Data
-  2  semantic     correction           computer-science / machine-learning  Batch Norm Placement Misconception
-  3  episodic     adaptation           physics / geophysics                 Gradient Explosion Under FP16
-
-Proceed with forking and PR creation? (y/n)
-```
-
-**STOP and wait for user confirmation.** Use `ask` (Codex) or print the table and end your turn so the user's next message is their response. Do not proceed without explicit consent.
-
-If the user wants to edit, adjust, or remove specific skills, do so before proceeding.
-
----
-
-## Stage 5 — Fork and Create PR
-
-### 5.1 — Check for existing fork
-
-```bash
-gh repo view "$(gh api user --jq '.login')/ResearchSkills" --json name 2>/dev/null
-```
-
-If no fork exists:
-
-```bash
-gh repo fork ScienceIntelligence/ResearchSkills --clone=false
-```
-
-### 5.2 — Clone the fork into a temp directory
-
-```bash
-WORK_DIR=$(mktemp -d)
-gh repo clone "$(gh api user --jq '.login')/ResearchSkills" "$WORK_DIR" -- --depth=1
-cd "$WORK_DIR"
-git remote add upstream https://github.com/ScienceIntelligence/ResearchSkills.git
-git fetch upstream main --depth=1
-git reset --hard upstream/main
-```
-
-### 5.3 — Create a branch
-
-```bash
-BRANCH="convert/$(gh api user --jq '.login')-$(date +%Y%m%d-%H%M%S)"
-git checkout -b "$BRANCH"
-```
-
-### 5.4 — Place skill files
-
-Each skill must be placed at the correct path. **Create directories that don't exist.**
-
-```
-.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
-```
-
-Rules:
-- `<domain>` — lowercase (e.g., `computer-science`)
-- `<subdomain>` — lowercase, hyphen-separated (e.g., `machine-learning`)
-- `<contributor>` — GitHub username from Stage 2
-- `<memory_type>` — `procedural`, `semantic`, or `episodic`
-- `<subtype>--<skill-name>` — lowercase, hyphen-separated (e.g., `correction--batch-norm-placement-misconception.md`)
-
-Example paths:
-```
-.agents/skills/statistics/methodology/jdoe/procedural/constraint-failure--auc-computation-with-masked-data.md
-.agents/skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement-misconception.md
-.agents/skills/physics/geophysics/jdoe/episodic/adaptation--gradient-explosion-under-fp16.md
-```
-
-For each skill:
-
-```bash
-SKILL_DIR=".agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>"
-mkdir -p "$SKILL_DIR"
-```
-
-Then write the skill file to `$SKILL_DIR/<subtype>--<skill-name>.md`.
-
-### 5.5 — Commit
-
-```bash
-git add .agents/skills/
-git commit -m "add(skills): N skills by <contributor>"
-```
-
-### 5.6 — Push and create PR
-
-```bash
-git push -u origin "$BRANCH"
-```
-
-Create a PR to the upstream repo:
-
-```bash
-gh pr create \
-  --repo ScienceIntelligence/ResearchSkills \
-  --head "$(gh api user --jq '.login'):$BRANCH" \
-  --title "[convert] Add N research skills by <contributor>" \
-  --body "$(cat <<'EOF'
-## Submission Type
-- [x] Research Skill (converted from existing notes/documents via `/researchskills-convert`)
-
-## Skills Added
-<!-- Fill this in: one line per skill, e.g. "- procedural / constraint-failure — statistics/methodology — AUC Computation With Masked Data" -->
-
-## Checklist
-- [x] Files placed in `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
-- [x] Filenames follow pattern `<subtype>--<skill-name>.md`
-- [x] All required frontmatter fields filled
-- [x] All required body sections present
-- [x] Content is de-identified
-- [x] All content in English
-EOF
-)"
-```
-
-### 5.7 — Clean up
-
-Only clean up after the PR was successfully created:
-
-```bash
-rm -rf "$WORK_DIR"
-```
-
-If any step in Stage 5 fails (fork, clone, push, or PR creation), show the error to the user and suggest manual steps to recover. Do not silently swallow errors.
-
----
-
-## Stage 6 — Report
-
-Show the user:
-
-```
-═══════════════════════════════════════════════════════
-  /researchskills-convert — Done!
-═══════════════════════════════════════════════════════
-
-  Submitted N skills via PR:
-    → <PR URL>
-
-  A domain reviewer will review your PR and merge it.
-  You can track the status on GitHub.
-═══════════════════════════════════════════════════════
-```
diff --git a/researchskills-extract/commands/SKILL.md b/researchskills-extract/commands/SKILL.md
deleted file mode 100644
index 39f8d16..0000000
--- a/researchskills-extract/commands/SKILL.md
+++ /dev/null
@@ -1,298 +0,0 @@
----
-name: "researchskills-extract"
-description: "Extract research skills from conversation history into ResearchSkills skill files."
----
-# /researchskills-extract
-
-Extract research skills from the user's Codex session history for **ResearchSkills**.
-
-**Run automatically with THREE pauses for user consent:** once after classifying projects (Stage 2.5 — choose which projects to scan), once after scoring to offer local installation (Stage 6.5 — store skills in Claude/Codex), and once before upload (Stage 7 — choose whether to submit). Report progress at each milestone.
-
-> **Prerequisite:** This skill spawns nested `codex exec` calls that need full network and filesystem access. Start Codex with: `codex -a never -s danger-full-access` (or `--dangerously-bypass-approvals-and-sandbox`). If the parent session is sandboxed, nested calls will fail with network errors.
-
-You extract three types of cognitive memory from research conversations:
-- **Procedural** — IF-THEN rules for **scientific research** decisions: methodology choices, data interpretation strategies, research direction pivots. NOT engineering workflows.
-- **Semantic** — **Frontier scientific knowledge** the LLM doesn't have: domain-specific constraints, unpublished findings, corrections to scientific misconceptions. NOT tool/API behaviors.
-- **Episodic** — **Research cognitive turning points**: hypothesis overturned, methodology abandoned for scientific reasons, unexpected findings that changed direction. NOT debugging episodes.
-
-Everything else — discovery, formatting, validation, upload — is done by helper scripts. Do not reimplement their work.
-
----
-
-## Pipeline
-
-```
-scan-sessions.js       ─┐
-classify-projects.js   ─┤
-extract-skills.js      ─┤  deterministic scripts (you call them)
-  └─ codex exec         │  ← Codex exec call per session, inside the script
-clean-skills.js        ─┤  ← review: reject/fix/merge
-score-skills.js        ─┤  ← score: 3-dim value assessment
-store-local.js         ─┤  ← optional: install skills into Claude/Codex
-finalize.js            ─┘
-
-You (main agent)       ← call scripts, read summaries, report
-```
-
-Helper scripts (installed at `~/.codex/skills/researchskills-extract/scripts/`):
-
-| Script | What it does |
-|--------|-------------|
-| `scan-sessions.js` | Discover sessions, extract metadata, filter, group by project |
-| `classify-projects.js` | Classify projects as research/engineering via Codex, pick domain/subdomain |
-| `extract-skills.js` | **The core loop**: format each session → call `codex exec` → validate + cache skills |
-| `validate-skills.js` | Validate skill markdown and cache to `~/.researchskills/cache/skills/` |
-| `clean-skills.js` | Review extracted skills: reject engineering, fix PII, merge duplicates |
-| `score-skills.js` | Score surviving skills on 3 dimensions: procedural, semantic, episodic value |
-| `store-local.js` | Install extracted skills into user's local Claude/Codex config |
-| `finalize.js` | Collect cached skills → upload to researchskills.ai |
-
----
-
-## Arguments
-
-- `--test` (alias: `test`): Test mode. Accept engineering sessions too (not just research). Tag all output as test data.
-- No argument: Production mode. Only research sessions proceed.
-
-Detect mode at start. Announce: `"Running in TEST MODE"` or `"Running in production mode"`.
-
----
-
-## Stage 1 — Scan
-
-```bash
-mkdir -p ~/.researchskills/cache/meta ~/.researchskills/cache/skills
-node ~/.codex/skills/researchskills-extract/scripts/scan-sessions.js
-```
-
-Reads `~/.researchskills/cache/work-list.json` output. Report: `"Found N sessions across M projects."`
-
----
-
-## Stage 2 — Classify Projects
-
-**YOU MUST call this script. Do NOT classify projects yourself.**
-
-```bash
-node ~/.codex/skills/researchskills-extract/scripts/classify-projects.js ~/.researchskills/cache/work-list.json --codex --verbose
-```
-
-For test mode, add `--test`.
-
-The script calls Codex to classify each project as research/engineering and picks domain/subdomain from the taxonomy. It also filters out non-research sessions (e.g., researchskills-extract runs, build/deploy tasks) via `skip_patterns`.
-
-Output: `~/.researchskills/cache/classification.json`.
-
-Read the output file. For each project with `type: "research"`, use its `research_session_ids` (NOT `session_ids`), `domain`, `subdomain`, and `project_name` in later stages. Do NOT include skipped sessions.
-
-The script generates an AI-summarized `project_name` for each project (e.g. "Protein Folding Simulation Pipeline") instead of using the raw folder name. Use this `project_name` in Stage 6 finalize. If `project_name` is null, fall back to the `slug`.
-
-Report: `"Classified N projects. Proceeding with M."`
-
----
-
-## Stage 2.5 — Project Consent Gate
-
-**PAUSE and ask the user.** After classification, show all discovered projects and let the user choose which to scan.
-
-Read `~/.researchskills/cache/classification.json` and display:
-
-```
-Select which projects to scan for research skills (all research projects selected — deselect any project that is not related to research):
-
-  [x] 1. Protein Folding Pipeline     (4 sessions, research, quantitative-biology)
-  [x] 2. Quantum Monte Carlo Study    (3 sessions, research, physics)
-  [ ] 3. Personal Website             (3 sessions, engineering)
-  [x] 4. Dotfiles                     (2 sessions, other)
-
-Enter numbers to toggle, or press Enter to continue:
-```
-
-All projects are pre-selected by default. Users can deselect individual projects by number.
-
-**YOU MUST STOP HERE AND WAIT FOR THE USER TO RESPOND.** Use `ask` (Codex) or `AskUserQuestion` (Claude Code) to present the project list and block until the user replies. Do NOT continue to Stage 3 without an explicit user response. If no interactive tool is available, print the list and end your turn — the user's next message is their selection.
-
-Only pass user-approved projects to Stage 3+. Remove deselected project session IDs from all subsequent `--session-ids` arguments.
-
-Report: `"Proceeding with N projects (M sessions) after user confirmation."`
-
----
-
-## Stage 3 — Extract Skills Per Session
-
-### MANDATORY: Use --single-batch and loop. NEVER run all at once.
-
-The extraction script MUST be called in a loop with `--single-batch`. Each call processes ONE batch (~5 parallel Codex calls) then exits. You call it again in a new tool call. This keeps the user informed of progress and prevents the UI from freezing.
-
-**FORBIDDEN patterns (will cause long freezes):**
-- `run_in_background: true` — user sees nothing for 10+ minutes
-- Omitting `--single-batch` — script runs all batches internally, no progress visible
-
-**REQUIRED pattern:**
-
-```bash
-# REPEAT this exact call in a loop. Each call = 1 batch.
-node ~/.codex/skills/researchskills-extract/scripts/extract-skills.js ~/.researchskills/cache/work-list.json \
-  --codex \
-  --domain <domain> \
-  --subdomain <subdomain> \
-  --contributor "$(git config user.name)" \
-  --session-ids <ALL-research-session-ids-csv> \
-  --single-batch \
-  --verbose
-```
-
-**Loop logic:**
-1. Run the command above (foreground, NOT background)
-2. Read the output. Report to user: "Batch N/M done: X skills extracted, Y calls remaining"
-3. If output says `0 Codex calls remaining` or `All sessions already cached` → **stop, go to Stage 4**
-4. Otherwise → run the **same command again** (it auto-skips cached segments)
-
-Pass ALL research session IDs from Stage 2. Do NOT drop sessions or pick a subset.
-
-If you need to process multiple projects with different domains, call the script once per project with `--session-ids` filtering to that project's sessions.
-
----
-
-## Stage 4 — Clean Skills
-
-Run review of all extracted skills: reject engineering content, fix PII leaks, merge duplicates.
-
-```bash
-node ~/.codex/skills/researchskills-extract/scripts/clean-skills.js \
-  --codex \
-  --session-ids <ALL-research-session-ids-csv> \
-  --verbose
-```
-
-This spawns a Codex instance that directly reads, deletes, and edits skill files on disk.
-
-Report: `"Clean: kept N, rejected M, merged K."`
-
----
-
-## Stage 5 — Score Skills
-
-Run assessment of the value of each surviving skill on 3 dimensions.
-
-```bash
-node ~/.codex/skills/researchskills-extract/scripts/score-skills.js \
-  --codex \
-  --session-ids <ALL-research-session-ids-csv> \
-  --verbose
-```
-
-This spawns a Codex instance that reads each skill and writes `review_scores` (procedural, semantic, episodic — each 0-5) into the YAML frontmatter.
-
-Report: `"Scored N skills. Avg: procedural X.X, semantic X.X, episodic X.X."`
-
----
-
-## Stage 6 — Finalize Per Project (collect only, no upload yet)
-
-Use the AI-generated `project_name` from classification.json (Stage 2). Do NOT use the raw folder name.
-
-**Do NOT pass `--upload` here.** Collect skills locally first. Upload requires explicit user consent in Stage 7.
-
-```bash
-node ~/.codex/skills/researchskills-extract/scripts/finalize.js \
-  --session-ids <ALL-research-session-ids-csv> \
-  --domain <domain> \
-  --subdomain <subdomain> \
-  --contributor "$(git config user.name)" \
-  --project-name "<project_name from classification>" \
-  --project-slug "<slug>"
-```
-
----
-
-## Stage 6.5 — Store Skills Locally (Optional)
-
-**Third consent gate.** After finalize collects skills, ask the user whether to install them into their local AI coding tool so the skills are available as context in future sessions.
-
-Use `ask` (Codex) or `AskUserQuestion` (Claude Code) to present the options. If no interactive tool is available, print the options and end your turn — the user's next message is their selection.
-
-- Question: "Install extracted skills into your local AI coding tool?"
-- Option A: "Yes, install to Claude Code" — stores skills to `~/.claude/commands/researchskills/<slug>.md`
-- Option B: "Yes, install to Codex" — stores skills to `~/.codex/skills/researchskills-<slug>/SKILL.md`
-- Option C: "Yes, install to both"
-- Option D: "No, skip local install"
-
-**YOU MUST STOP HERE AND WAIT FOR THE USER TO RESPOND.** Do NOT continue to Stage 7 without an explicit user response.
-
-If the user picks A, B, or C, run:
-
-```bash
-node ~/.codex/skills/researchskills-extract/scripts/store-local.js \
-  --target <claude|codex|both> \
-  --session-ids <ALL-research-session-ids-csv>
-```
-
-Report: `"Installed N skills to <target>. M already up-to-date."`
-
-If the user picks D, skip and continue to Stage 7.
-
----
-
-## Stage 7 — Consent and Upload
-
-**Fourth consent gate.** Pause and ask the user before uploading anything.
-
-Show the user what was extracted:
-
-```
-═══════════════════════════════════════════════════════
-  /researchskills-extract — Extraction Complete!
-═══════════════════════════════════════════════════════
-
-Extracted N skills from M sessions across P projects:
-  • Episodic:   E skills
-  • Semantic:   S skills
-  • Procedural: Pr skills
-
-Review:
-  • Kept: K / Rejected: R / Merged: G
-  • Avg scores: procedural X.X, semantic X.X, episodic X.X
-
-⚠ Nothing has been uploaded yet. Your skills are saved
-  locally. Would you like to submit them to ResearchSkills
-  for reviewer review?
-
-  Skills will be stored on researchskills.ai and reviewed
-  by a maintainer before publication (CC-BY 4.0).
-═══════════════════════════════════════════════════════
-```
-
-Ask for explicit consent:
-- "Yes, submit for review" — re-run finalize with `--upload`
-- "No, keep local only" — skip upload, tell user where files are saved
-
-If the user consents, re-run finalize with `--upload`.
-
-**Headless/SSH detection:** If running over SSH (SSH_CONNECTION or SSH_CLIENT env vars set) or on a headless Linux server (no DISPLAY), the upload script automatically detects this and:
-- Disables browser opening
-- Prints the review URL for the user to visit from any browser
-
-When the user has consented via the prompt above, pass `--consent` to include `consent: true` in the upload payload.
-
-```bash
-node ~/.codex/skills/researchskills-extract/scripts/finalize.js \
-  --session-ids <ALL-research-session-ids-csv> \
-  --domain <domain> \
-  --subdomain <subdomain> \
-  --contributor "$(git config user.name)" \
-  --project-name "<project_name from classification>" \
-  --project-slug "<slug>" \
-  --upload
-```
-
-Then show:
-```
-Review your skills:
-  → https://researchskills.ai/review/batch/<batchId>
-```
-
-If headless, also show:
-```
-  (Sign in with GitHub on the review page to claim credit and submit.)
-```
diff --git a/researchskills-extract/commands/researchskills-convert.md b/researchskills-extract/commands/researchskills-convert.md
deleted file mode 100644
index 298e483..0000000
--- a/researchskills-extract/commands/researchskills-convert.md
+++ /dev/null
@@ -1,268 +0,0 @@
-# /researchskills-convert
-
-Convert existing research skills from any format (notes, documents, other AI tools) into **ResearchSkills** format, then fork the repo and open a PR — all in one step.
-
-**This is an interactive skill.** Ask the user ONE question at a time. Do not dump all questions at once.
-
----
-
-## Stage 1 — Collect Input
-
-Ask the user:
-
-> **Where are your skills?**
-> Provide a file path, a directory, or paste them here.
-
-Read the files or accept the pasted text. If a directory, recursively read all `.md`, `.txt`, `.json`, and `.yaml` files in it.
-
----
-
-## Stage 2 — Identify the User
-
-Get the GitHub username for the `contributor` field:
-
-```bash
-gh api user --jq '.login'
-```
-
-If `gh` is not installed or not authenticated, tell the user:
-
-> `gh` (GitHub CLI) is required for PR creation. Install it:
-> - macOS: `brew install gh`
-> - Linux: see https://cli.github.com
->
-> Then run `gh auth login` to authenticate.
-
-**Stop here until `gh api user` succeeds.**
-
----
-
-## Stage 3 — Convert Skills
-
-Analyze the input and convert each genuine **research** skill into the ResearchSkills format. Apply these filters:
-
-### What to INCLUDE
-- Domain-specific research insights a researcher would find non-obvious
-- Methodology decisions, scientific corrections, experimental lessons
-- Knowledge that is frontier, non-public, or corrects LLM misconceptions
-
-### What to SKIP
-- Generic software engineering (git workflows, CI/CD, Docker, deployment)
-- DevOps, infrastructure, or tooling knowledge
-- Textbook material readily available in LLM training data
-- Content that is not related to scientific research
-
-### Memory Types and Required Sections
-
-**Procedural** (subtypes: `tie`, `no-change`, `constraint-failure`, `operator-fail`)
-- When — trigger conditions and exclusions
-- Decision — Preferred, Rejected, Reasoning
-- Local Verifiers — concrete diagnostics
-- Failure Handling — fallback strategies
-- Anti-exemplars — when NOT to use (recommended)
-
-**Semantic** (subtypes: `frontier`, `non-public`, `correction`)
-- Fact — precise core claim
-- Evidence — how you know this is true
-- LLM Default Belief — **correction subtype only**, delete for others
-- Expiry Signal — when to revisit
-
-**Episodic** (subtypes: `failure`, `adaptation`, `anomalous`)
-- Situation — tools, dataset, parameters, expectations
-- Action — what was done, with specifics
-- Outcome — concrete result with metrics
-- Lesson — specific IF-THEN rule
-- Retrieval Cues — trigger conditions for recall
-
-### Frontmatter
-
-Every skill file must have this exact frontmatter:
-
-```yaml
----
-name: "Skill Name In Title Case"
-memory_type: procedural          # procedural | semantic | episodic
-subtype: tie                     # see subtypes above
-domain: computer-science         # see domain list below
-subdomain: machine-learning      # arXiv-aligned subdomain
-contributor: gh-username          # from Stage 2
----
-```
-
-### Valid Domains
-
-Pick from these arXiv-aligned domains:
-- `physics`
-- `mathematics`
-- `computer-science`
-- `quantitative-biology`
-- `statistics`
-- `eess` (electrical engineering and systems science)
-- `economics`
-- `quantitative-finance`
-
-For subdomain, use an arXiv-aligned subcategory (e.g., `machine-learning`, `geophysics`, `genomics`, `methodology`).
-
-### De-identification
-
-Before outputting, strip:
-- Personal names (replace with role descriptions)
-- Private file paths (use generic paths)
-- Internal URLs and hostnames
-- Lab-specific identifiers
-- API keys, tokens, credentials
-
-All output MUST be in English. If source material is in another language, paraphrase in English.
-
----
-
-## Stage 4 — Show Preview
-
-Show the user all converted skills in a summary table:
-
-```
-Converted N skills:
-
-  #  Type         Subtype              Domain / Subdomain                   Name
-  1  procedural   constraint-failure   statistics / methodology             AUC Computation With Masked Data
-  2  semantic     correction           computer-science / machine-learning  Batch Norm Placement Misconception
-  3  episodic     adaptation           physics / geophysics                 Gradient Explosion Under FP16
-
-Proceed with forking and PR creation? (y/n)
-```
-
-**STOP and wait for user confirmation.** Use AskUserQuestion (Claude Code) to present the table and block until the user replies. Do not proceed without explicit consent.
-
-If the user wants to edit, adjust, or remove specific skills, do so before proceeding.
-
----
-
-## Stage 5 — Fork and Create PR
-
-### 5.1 — Check for existing fork
-
-```bash
-gh repo view "$(gh api user --jq '.login')/ResearchSkills" --json name 2>/dev/null
-```
-
-If no fork exists:
-
-```bash
-gh repo fork ScienceIntelligence/ResearchSkills --clone=false
-```
-
-### 5.2 — Clone the fork into a temp directory
-
-```bash
-WORK_DIR=$(mktemp -d)
-gh repo clone "$(gh api user --jq '.login')/ResearchSkills" "$WORK_DIR" -- --depth=1
-cd "$WORK_DIR"
-git remote add upstream https://github.com/ScienceIntelligence/ResearchSkills.git
-git fetch upstream main --depth=1
-git reset --hard upstream/main
-```
-
-### 5.3 — Create a branch
-
-```bash
-BRANCH="convert/$(gh api user --jq '.login')-$(date +%Y%m%d-%H%M%S)"
-git checkout -b "$BRANCH"
-```
-
-### 5.4 — Place skill files
-
-Each skill must be placed at the correct path. **Create directories that don't exist.**
-
-```
-.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
-```
-
-Rules:
-- `<domain>` — lowercase (e.g., `computer-science`)
-- `<subdomain>` — lowercase, hyphen-separated (e.g., `machine-learning`)
-- `<contributor>` — GitHub username from Stage 2
-- `<memory_type>` — `procedural`, `semantic`, or `episodic`
-- `<subtype>--<skill-name>` — lowercase, hyphen-separated (e.g., `correction--batch-norm-placement-misconception.md`)
-
-Example paths:
-```
-.agents/skills/statistics/methodology/jdoe/procedural/constraint-failure--auc-computation-with-masked-data.md
-.agents/skills/computer-science/machine-learning/jdoe/semantic/correction--batch-norm-placement-misconception.md
-.agents/skills/physics/geophysics/jdoe/episodic/adaptation--gradient-explosion-under-fp16.md
-```
-
-For each skill:
-
-```bash
-SKILL_DIR=".agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>"
-mkdir -p "$SKILL_DIR"
-```
-
-Then write the skill file to `$SKILL_DIR/<subtype>--<skill-name>.md`.
-
-### 5.5 — Commit
-
-```bash
-git add .agents/skills/
-git commit -m "add(skills): N skills by <contributor>"
-```
-
-### 5.6 — Push and create PR
-
-```bash
-git push -u origin "$BRANCH"
-```
-
-Create a PR to the upstream repo:
-
-```bash
-gh pr create \
-  --repo ScienceIntelligence/ResearchSkills \
-  --head "$(gh api user --jq '.login'):$BRANCH" \
-  --title "[convert] Add N research skills by <contributor>" \
-  --body "$(cat <<'EOF'
-## Submission Type
-- [x] Research Skill (converted from existing notes/documents via `/researchskills-convert`)
-
-## Skills Added
-<!-- Fill this in: one line per skill, e.g. "- procedural / constraint-failure — statistics/methodology — AUC Computation With Masked Data" -->
-
-## Checklist
-- [x] Files placed in `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/`
-- [x] Filenames follow pattern `<subtype>--<skill-name>.md`
-- [x] All required frontmatter fields filled
-- [x] All required body sections present
-- [x] Content is de-identified
-- [x] All content in English
-EOF
-)"
-```
-
-### 5.7 — Clean up
-
-Only clean up after the PR was successfully created:
-
-```bash
-rm -rf "$WORK_DIR"
-```
-
-If any step in Stage 5 fails (fork, clone, push, or PR creation), show the error to the user and suggest manual steps to recover. Do not silently swallow errors.
-
----
-
-## Stage 6 — Report
-
-Show the user:
-
-```
-═══════════════════════════════════════════════════════
-  /researchskills-convert — Done!
-═══════════════════════════════════════════════════════
-
-  Submitted N skills via PR:
-    → <PR URL>
-
-  A domain reviewer will review your PR and merge it.
-  You can track the status on GitHub.
-═══════════════════════════════════════════════════════
-```
diff --git a/researchskills-extract/commands/researchskills-extract.md b/researchskills-extract/commands/researchskills-extract.md
deleted file mode 100644
index e064df5..0000000
--- a/researchskills-extract/commands/researchskills-extract.md
+++ /dev/null
@@ -1,294 +0,0 @@
-# /researchskills-extract
-
-Extract research skills from the user's Claude Code session history for **ResearchSkills**.
-
-**Run automatically with THREE pauses for user consent:** once after classifying projects (Stage 2.5 — choose which projects to scan), once after scoring to offer local installation (Stage 6.5 — store skills in Claude/Codex), and once before upload (Stage 7 — choose whether to submit). Report progress at each milestone.
-
-You extract three types of cognitive memory from research conversations:
-- **Procedural** — IF-THEN rules for **scientific research** decisions: methodology choices, data interpretation strategies, research direction pivots. NOT engineering workflows.
-- **Semantic** — **Frontier scientific knowledge** the LLM doesn't have: domain-specific constraints, unpublished findings, corrections to scientific misconceptions. NOT tool/API behaviors.
-- **Episodic** — **Research cognitive turning points**: hypothesis overturned, methodology abandoned for scientific reasons, unexpected findings that changed direction. NOT debugging episodes.
-
-Everything else — discovery, formatting, validation, upload — is done by helper scripts. Do not reimplement their work.
-
----
-
-## Pipeline
-
-```
-scan-sessions.js       ─┐
-classify-projects.js   ─┤
-extract-skills.js      ─┤  deterministic scripts (you call them)
-  └─ claude -p sonnet   │  ← Sonnet CLI call per session, inside the script
-clean-skills.js        ─┤  ← Opus reviews: reject/fix/merge
-score-skills.js        ─┤  ← Opus scores: 3-dim value assessment
-store-local.js         ─┤  ← optional: install skills into Claude/Codex
-finalize.js            ─┘
-
-You (main agent)       ← call scripts, read summaries, report
-```
-
-Helper scripts (installed at `~/.claude/utils/`):
-
-| Script | What it does |
-|--------|-------------|
-| `scan-sessions.js` | Discover sessions, extract metadata, filter, group by project |
-| `classify-projects.js` | Classify projects as research/engineering via Sonnet, pick domain/subdomain |
-| `extract-skills.js` | **The core loop**: format each session → call `claude -p --model sonnet` → validate + cache skills |
-| `validate-skills.js` | Validate skill markdown and cache to `~/.researchskills/cache/skills/` |
-| `clean-skills.js` | Review extracted skills with Opus: reject engineering, fix PII, merge duplicates |
-| `score-skills.js` | Score surviving skills with Opus on 3 dimensions: procedural, semantic, episodic value |
-| `store-local.js` | Install extracted skills into user's local Claude/Codex config |
-| `finalize.js` | Collect cached skills → upload to researchskills.ai |
-
----
-
-## Arguments
-
-- `--test` (alias: `test`): Test mode. Accept engineering sessions too (not just research). Tag all output as test data.
-- No argument: Production mode. Only research sessions proceed.
-
-Detect mode at start. Announce: `"Running in TEST MODE"` or `"Running in production mode"`.
-
----
-
-## Stage 1 — Scan
-
-```bash
-mkdir -p ~/.researchskills/cache/meta ~/.researchskills/cache/skills
-node ~/.claude/utils/scan-sessions.js
-```
-
-Reads `~/.researchskills/cache/work-list.json` output. Report: `"Found N sessions across M projects."`
-
----
-
-## Stage 2 — Classify Projects
-
-**YOU MUST call this script. Do NOT classify projects yourself.**
-
-```bash
-node ~/.claude/utils/classify-projects.js ~/.researchskills/cache/work-list.json --cc --verbose
-```
-
-For test mode, add `--test`.
-
-The script calls Sonnet to classify each project as research/engineering and picks domain/subdomain from the taxonomy. It also filters out non-research sessions (e.g., researchskills-extract runs, build/deploy tasks) via `skip_patterns`.
-
-Output: `~/.researchskills/cache/classification.json`.
-
-Read the output file. For each project with `type: "research"`, use its `research_session_ids` (NOT `session_ids`), `domain`, `subdomain`, and `project_name` in later stages. Do NOT include skipped sessions.
-
-The script generates an AI-summarized `project_name` for each project (e.g. "Protein Folding Simulation Pipeline") instead of using the raw folder name. Use this `project_name` in Stage 6 finalize. If `project_name` is null, fall back to the `slug`.
-
-Report: `"Classified N projects. Proceeding with M."`
-
----
-
-## Stage 2.5 — Project Consent Gate
-
-**PAUSE and ask the user.** After classification, show all discovered projects and let the user choose which to scan.
-
-Read `~/.researchskills/cache/classification.json` and display:
-
-```
-Select which projects to scan for research skills (all research projects selected — deselect any project that is not related to research):
-
-  [x] 1. Protein Folding Pipeline     (4 sessions, research, quantitative-biology)
-  [x] 2. Quantum Monte Carlo Study    (3 sessions, research, physics)
-  [ ] 3. Personal Website             (3 sessions, engineering)
-  [x] 4. Dotfiles                     (2 sessions, other)
-
-Enter numbers to toggle, or press Enter to continue:
-```
-
-All research projects are pre-selected by default. Users can deselect individual projects by number.
-
-**YOU MUST STOP HERE AND WAIT FOR THE USER TO RESPOND.** Use AskUserQuestion to present the project list and block until the user replies. Do NOT continue to Stage 3 without an explicit user response.
-
-Only pass user-approved projects to Stage 3+. Remove deselected project session IDs from all subsequent `--session-ids` arguments.
-
-Report: `"Proceeding with N projects (M sessions) after user confirmation."`
-
----
-
-## Stage 3 — Extract Skills Per Session
-
-### MANDATORY: Use --single-batch and loop. NEVER run all at once.
-
-The extraction script MUST be called in a loop with `--single-batch`. Each call processes ONE batch (~5 parallel Sonnet calls) then exits. You call it again in a new Bash tool call. This keeps the user informed of progress and prevents the UI from freezing.
-
-**FORBIDDEN patterns (will cause long freezes):**
-- `run_in_background: true` — user sees nothing for 10+ minutes
-- Omitting `--single-batch` — script runs all batches internally, no progress visible
-- Using Monitor tool to watch output — still freezes, just with delayed notifications
-
-**REQUIRED pattern:**
-
-```bash
-# REPEAT this exact Bash call in a loop. Each call = 1 batch.
-node ~/.claude/utils/extract-skills.js ~/.researchskills/cache/work-list.json \
-  --cc \
-  --domain <domain> \
-  --subdomain <subdomain> \
-  --contributor "$(git config user.name)" \
-  --session-ids <ALL-research-session-ids-csv> \
-  --single-batch \
-  --verbose
-```
-
-**Loop logic:**
-1. Run the command above (foreground Bash, NOT background)
-2. Read the output. Report to user: "Batch N/M done: X skills extracted, Y calls remaining"
-3. If output says `0 Sonnet calls remaining` or `All sessions already cached` → **stop, go to Stage 4**
-4. Otherwise → run the **same command again** (it auto-skips cached segments)
-
-Pass ALL research session IDs from Stage 2. Do NOT drop sessions or pick a subset.
-
-If you need to process multiple projects with different domains, call the script once per project with `--session-ids` filtering to that project's sessions.
-
----
-
-## Stage 4 — Clean Skills
-
-Run Opus to review all extracted skills: reject engineering content, fix PII leaks, merge duplicates.
-
-```bash
-node ~/.claude/utils/clean-skills.js \
-  --cc \
-  --session-ids <ALL-research-session-ids-csv> \
-  --verbose
-```
-
-This spawns a Claude Code instance with Opus that directly reads, deletes, and edits skill files on disk.
-
-Report: `"Clean: kept N, rejected M, merged K."`
-
----
-
-## Stage 5 — Score Skills
-
-Run Opus to assess the value of each surviving skill on 3 dimensions.
-
-```bash
-node ~/.claude/utils/score-skills.js \
-  --cc \
-  --session-ids <ALL-research-session-ids-csv> \
-  --verbose
-```
-
-This spawns a Claude Code instance with Opus that reads each skill and writes `review_scores` (procedural, semantic, episodic — each 0-5) into the YAML frontmatter.
-
-Report: `"Scored N skills. Avg: procedural X.X, semantic X.X, episodic X.X."`
-
----
-
-## Stage 6 — Finalize Per Project (collect only, no upload yet)
-
-Use the AI-generated `project_name` from classification.json (Stage 2). Do NOT use the raw folder name.
-
-**Do NOT pass `--upload` here.** Collect skills locally first. Upload requires explicit user consent in Stage 7.
-
-```bash
-node ~/.claude/utils/finalize.js \
-  --session-ids <ALL-research-session-ids-csv> \
-  --domain <domain> \
-  --subdomain <subdomain> \
-  --contributor "$(git config user.name)" \
-  --project-name "<project_name from classification>" \
-  --project-slug "<slug>"
-```
-
----
-
-## Stage 6.5 — Store Skills Locally (Optional)
-
-**Third consent gate.** After finalize collects skills, ask the user whether to install them into their local AI coding tool so the skills are available as context in future sessions.
-
-Use AskUserQuestion to present the options:
-- Question: "Install extracted skills into your local AI coding tool?"
-- Option A: "Yes, install to Claude Code" — stores skills to `~/.claude/commands/researchskills/<slug>.md`
-- Option B: "Yes, install to Codex" — stores skills to `~/.codex/skills/researchskills-<slug>/SKILL.md`
-- Option C: "Yes, install to both"
-- Option D: "No, skip local install"
-
-**YOU MUST STOP HERE AND WAIT FOR THE USER TO RESPOND.** Do NOT continue to Stage 7 without an explicit user response.
-
-If the user picks A, B, or C, run:
-
-```bash
-node ~/.claude/utils/store-local.js \
-  --target <claude|codex|both> \
-  --session-ids <ALL-research-session-ids-csv>
-```
-
-Report: `"Installed N skills to <target>. M already up-to-date."`
-
-If the user picks D, skip and continue to Stage 7.
-
----
-
-## Stage 7 — Consent and Upload
-
-**Fourth consent gate.** Pause and ask the user before uploading anything.
-
-Show the user what was extracted:
-
-```
-═══════════════════════════════════════════════════════
-  /researchskills-extract — Extraction Complete!
-═══════════════════════════════════════════════════════
-
-Extracted N skills from M sessions across P projects:
-  • Episodic:   E skills
-  • Semantic:   S skills
-  • Procedural: Pr skills
-
-Review (Opus):
-  • Kept: K / Rejected: R / Merged: G
-  • Avg scores: procedural X.X, semantic X.X, episodic X.X
-
-⚠ Nothing has been uploaded yet. Your skills are saved
-  locally. Would you like to submit them to ResearchSkills
-  for reviewer review?
-
-  Skills will be stored on researchskills.ai and reviewed
-  by a maintainer before publication (CC-BY 4.0).
-═══════════════════════════════════════════════════════
-```
-
-Then use AskUserQuestion to get explicit consent:
-- Question: "Submit your extracted skills to ResearchSkills for review?"
-- Option A: "Yes, submit for review" — re-run finalize with `--upload`
-- Option B: "No, keep local only" — skip upload, tell user where files are saved
-
-If the user consents, re-run finalize with `--upload`.
-
-**Headless/SSH detection:** If running over SSH (SSH_CONNECTION or SSH_CLIENT env vars set) or on a headless Linux server (no DISPLAY), the upload script automatically detects this and:
-- Disables browser opening
-- Prints the review URL for the user to visit from any browser
-
-When the user has consented via the prompt above, pass `--consent` to include `consent: true` in the upload payload.
-
-```bash
-node ~/.claude/utils/finalize.js \
-  --session-ids <ALL-research-session-ids-csv> \
-  --domain <domain> \
-  --subdomain <subdomain> \
-  --contributor "$(git config user.name)" \
-  --project-name "<project_name from classification>" \
-  --project-slug "<slug>" \
-  --upload
-```
-
-Then show:
-```
-Review your skills:
-  → https://researchskills.ai/review/batch/<batchId>
-```
-
-If headless, also show:
-```
-  (Sign in with GitHub on the review page to claim credit and submit.)
-```
-
diff --git a/researchskills-extract/docs/codex-migration-plan.md b/researchskills-extract/docs/codex-migration-plan.md
deleted file mode 100644
index d54f418..0000000
--- a/researchskills-extract/docs/codex-migration-plan.md
+++ /dev/null
@@ -1,400 +0,0 @@
-# Codex Migration Plan
-
-> 从 Claude Code 迁移 researchskills-extract 到 Codex 的平台差异参考。
-> 内容与具体 JS 实现正交——无论 skills 脚本怎么改，这些平台事实不变。
-
----
-
-## 1. Codex Session 文件格式（实测验证 2026-04）
-
-### 文件位置
-
-```
-~/.codex/sessions/YYYY/MM/DD/rollout-<datetime>-<uuid>.jsonl
-~/.codex/archived_sessions/rollout-<datetime>-<uuid>.jsonl
-~/.codex/session_index.jsonl   # 索引：{id, thread_name, updated_at}
-```
-
-### JSONL 事件类型全表
-
-每行一个 JSON 对象，`type` 字段区分事件类型，数据在 `payload` 中。
-
-#### session_meta（首行，每个文件只有一条）
-
-```json
-{
-  "timestamp": "2026-03-12T13:29:01.469Z",
-  "type": "session_meta",
-  "payload": {
-    "id": "019ce23c-0ff4-7fb1-aee9-a3bb96504102",
-    "cwd": "/Users/genghejia/Desktop/ai_lawyer",
-    "originator": "codex_vscode",
-    "cli_version": "0.108.0-alpha.12",
-    "source": "vscode",
-    "model_provider": "openai",
-    "base_instructions": { "text": "..." }
-  }
-}
-```
-
-关键字段：`payload.id`（session UUID）、`payload.cwd`（项目路径）
-
-#### event_msg — 用户消息
-
-```json
-{
-  "timestamp": "...",
-  "type": "event_msg",
-  "payload": {
-    "type": "user_message",
-    "message": "用户输入的文本",
-    "images": [],
-    "local_images": [],
-    "text_elements": []
-  }
-}
-```
-
-#### event_msg — Agent 评论
-
-```json
-{
-  "type": "event_msg",
-  "payload": {
-    "type": "agent_message",
-    "message": "Agent 的解释性文本...",
-    "phase": "commentary"
-  }
-}
-```
-
-`phase` 可能值：`"commentary"`, `"final_answer"`
-
-#### event_msg — 其他（可跳过）
-
-| `payload.type` | 说明 | 提取价值 |
-|---|---|---|
-| `token_count` | token 使用量统计 | 无，跳过 |
-| `task_started` | 任务开始，含 turn_id | 可用于时间戳 |
-| `task_complete` | 任务结束，含 last_agent_message | 可用于摘要 |
-
-#### response_item — 工具调用
-
-```json
-{
-  "type": "response_item",
-  "payload": {
-    "type": "function_call",
-    "name": "exec_command",
-    "arguments": "{\"cmd\":\"pwd && ls\",\"workdir\":\"/path\",\"max_output_tokens\":4000}",
-    "call_id": "call_Vl7nOpPrsZRGrGi42QDkNCwJ"
-  }
-}
-```
-
-工具名：`exec_command`（shell 命令）、`write_stdin`
-
-#### response_item — 工具输出
-
-```json
-{
-  "type": "response_item",
-  "payload": {
-    "type": "function_call_output",
-    "call_id": "call_Vl7nOpPrsZRGrGi42QDkNCwJ",
-    "output": "Chunk ID: f896ae\nWall time: 0.0509 seconds\nProcess exited with code 0\nOriginal token count: 2571\nOutput:\n实际输出内容..."
-  }
-}
-```
-
-注意：output 包含元数据前缀（Chunk ID, Wall time, exit code, token count），实际内容在 `Output:\n` 之后。
-
-#### response_item — apply_patch（custom_tool_call）
-
-```json
-{
-  "type": "response_item",
-  "payload": {
-    "type": "custom_tool_call",
-    "status": "completed",
-    "call_id": "call_xxx",
-    "name": "apply_patch",
-    "input": "*** Begin Patch\n*** Add File: /path/to/file.js\n+line1\n+line2\n*** End Patch"
-  }
-}
-```
-
-#### response_item — Assistant 消息
-
-```json
-{
-  "type": "response_item",
-  "payload": {
-    "type": "message",
-    "role": "assistant",
-    "content": [{ "type": "output_text", "text": "..." }],
-    "phase": "commentary"
-  }
-}
-```
-
-#### response_item — 推理（加密，无法读取）
-
-```json
-{
-  "type": "response_item",
-  "payload": {
-    "type": "reasoning",
-    "encrypted_content": "gAAAAAB...(base64)..."
-  }
-}
-```
-
-跳过，加密内容无法解密。
-
-#### turn_context（每个 turn 开头）
-
-```json
-{
-  "type": "turn_context",
-  "payload": {
-    "turn_id": "...",
-    "cwd": "/path",
-    "model": "gpt-5.4",
-    "personality": "pragmatic",
-    "effort": "low"
-  }
-}
-```
-
-对提取无直接价值，但 `model` 字段可用于判断用了什么模型。
-
----
-
-## 2. Claude Code vs Codex 对照表
-
-### 消息结构
-
-| 概念 | Claude Code | Codex |
-|------|------------|-----------|
-| 用户消息 | `{type:"user", message:{role:"user", content}}` | `{type:"event_msg", payload:{type:"user_message", message}}` |
-| Assistant 消息 | `{type:"assistant", message:{role:"assistant", content}}` | `{type:"response_item", payload:{type:"message", role:"assistant", content:[{type:"output_text", text}]}}` 或 `{type:"event_msg", payload:{type:"agent_message", message}}` |
-| 工具调用 | content 数组中的 `{type:"tool_use", name, input}` | 独立行 `{type:"response_item", payload:{type:"function_call", name, arguments, call_id}}` |
-| 工具结果 | content 数组中的 `{type:"tool_result", content}` | 独立行 `{type:"response_item", payload:{type:"function_call_output", call_id, output}}` |
-| 文件编辑 | `Edit` 工具 (old_string/new_string) | `apply_patch` (custom_tool_call, unified diff) |
-| 时间戳 | 顶层 `timestamp` 字段 | 顶层 `timestamp` 字段 |
-| Session ID | 文件名即 ID（`<uuid>.jsonl`） | 文件名中的 UUID 或 `session_meta.payload.id` |
-| 项目路径 | JSONL 中的 `cwd` 或目录名编码 | `session_meta.payload.cwd` |
-
-### 工具体系
-
-| Claude Code 工具 | Codex 等价 | 压缩策略建议 |
-|---|---|---|
-| `Read` | `exec_command` → `cat/head/tail` | 丢弃输出，只保留文件名 |
-| `Edit` / `Write` | `apply_patch` | 只保留文件名，保留 rejection |
-| `Bash` | `exec_command` | 同 CC：错误 120 字符，正常 60 字符 |
-| `Grep` | `exec_command` → `grep/rg` | 丢弃 |
-| `Glob` | `exec_command` → `find/fd/ls` | 丢弃 |
-| `Agent` | 无直接等价 | N/A |
-| `TodoWrite` | 无直接等价 | N/A |
-
-Codex `exec_command` 需要解析 `cmd` 字符串来判断意图：
-```
-cat/head/tail/less  → 文件读取
-grep/rg/ag          → 搜索
-find/fd/ls          → 文件发现
-git *               → git 操作（��弃）
-其他                → 通用 shell 命令
-```
-
-### 丢弃/跳过的事件
-
-| 事件 | 原因 |
-|------|------|
-| `session_meta` | 元数据，非对话内容（但需提取 cwd） |
-| `token_count` | 计费信息，无研究价值 |
-| `task_started` / `task_complete` | 流程控制，非知识 |
-| `reasoning` | 加密，无法读取 |
-| `turn_context` | 配置上下文，非对话 |
-
-### 安装路径
-
-| 用途 | Claude Code | Codex |
-|------|------------|-----------|
-| Skill/Command 文件 | `~/.claude/commands/researchskills-extract.md` | `~/.codex/skills/researchskills-extract/SKILL.md` |
-| 辅助脚本 | `~/.claude/utils/` | `~/.codex/skills/researchskills-extract/scripts/` |
-| Session 数据 | `~/.claude/projects/` | `~/.codex/sessions/` + `~/.codex/archived_sessions/` |
-| 共享缓存 | `~/.researchskills/cache/` | `~/.researchskills/cache/`（共用） |
-
----
-
-## 3. Codex 非交互模式（`codex exec`）
-
-用于 AI 提取阶段，等价于 Claude Code 的 `claude -p --model haiku`。
-
-```bash
-# 基本用法：prompt 通过 stdin 传入
-echo "$PROMPT" | codex exec --full-auto --ephemeral
-
-# 指定模型
-echo "$PROMPT" | codex exec --full-auto --ephemeral -m <model>
-
-# JSON 输出（结构化）
-echo "$PROMPT" | codex exec --full-auto --ephemeral --json
-
-# 输出到文件
-codex exec --full-auto --ephemeral -o /tmp/result.txt "prompt here"
-```
-
-关键 flag：
-
-| Flag | 作用 |
-|------|------|
-| `--full-auto` | 自动批准所有操作，无交互 |
-| `--ephemeral` | 不持久化 session（避免污染用户 session 历史） |
-| `-m <model>` | 指定模型 |
-| `--json` | JSONL 事件流输出 |
-| `-o <file>` | 最终消息写入文件 |
-| `-C <dir>` | 设置工作目录 |
-| `--skip-git-repo-check` | 允许在非 git 目录运行 |
-
-**输出行为**：
-- 默认：进度信息 → stderr，最终回答 → stdout
-- `--json`：所有事件以 JSONL 流式输出到 stdout
-
----
-
-## 4. SKILL.md 格式差异
-
-### Claude Code（当前）
-
-`~/.claude/commands/researchskills-extract.md` — 纯 Markdown，无需 frontmatter：
-
-```markdown
-# /researchskills-extract
-
-Extract research skills from...
-```
-
-### Codex（需要 YAML frontmatter）
-
-`~/.codex/skills/researchskills-extract/SKILL.md`：
-
-```yaml
----
-name: "researchskills-extract"
-description: "Extract research skills from conversation history into ResearchSkills skill files."
----
-# /researchskills-extract
-
-Extract research skills from...
-```
-
-Codex 只在启动时读 frontmatter（name + description），正文在触发后才加载。
-
-### 跨平台建议
-
-SKILL.md 正文内容可以完全相同。postinstall 时：
-- Claude Code：直接复制
-- Codex：在头部加 YAML frontmatter 后复制
-
----
-
-## 5. Adapter 架构设计
-
-### 接口
-
-每个平台 adapter 导出一个普通对象，实现以下方法：
-
-```js
-{
-  name: 'claude-code' | 'codex-cli',
-
-  // Session 发现
-  discoverSessions(),                  // → string[]
-  extractSessionId(filePath),          // → string
-  deriveProjectPath(filePath, meta),   // → string
-
-  // JSONL 解析
-  parseEntry(jsonParsedLine),          // → { role, text, timestamp, tools[] } | null
-  isDroppedTool(toolName),             // → boolean
-  compressTool(toolName, toolInfo, output), // → string | null
-
-  // AI 调用
-  runExtraction(prompt, timeoutMs),    // → Promise<{ ok, output, error }>
-}
-```
-
-### 平台检测
-
-优先级：
-1. `--platform` CLI 参数（显式覆盖）
-2. 环境变量 `EXTRACT_KNOWHOW_PLATFORM`
-3. 文件系统检测（哪个目录有 session 数据）
-4. 两个都有 → 两个都扫（合并结果，`source` 字段区分）
-
-### 共享逻辑（不在 adapter 中）
-
-以下逻辑与平台无关，保留在主脚本中：
-- Session 过滤（文件大小、消息数、时长、subagent、去重）
-- format-session 的 Pass 1.5（清理用户消息）、Pass 3（折叠连续工具）、Pass 4（过滤噪音）
-- 验证（validate-skills.js）
-- 上传（upload-skills.js）
-- 打包（finalize.js）
-- 提取 prompt 模板
-- 缓存机制
-
----
-
-## 6. Codex exec_command 输出解析
-
-Codex 的 `function_call_output` 有固定的元数据前缀格式：
-
-```
-Chunk ID: f896ae
-Wall time: 0.0509 seconds
-Process exited with code 0
-Original token count: 2571
-Output:
-<实际命令输出>
-```
-
-解析时需要：
-1. 提取 `Process exited with code N` 判断成功/失败
-2. 跳过前缀行，从 `Output:\n` 之后开始读实际内容
-3. 用 exit code + 输出内容中的关键词（Error, Traceback, panic 等）判断是否为错误输出
-
----
-
-## 7. Subagent 检测
-
-### Claude Code
-- Subagent session 在 `subagents/` 子目录中
-- 或首 5 行包含 `RESPOND WITH ONLY A VALID JSON OBJECT` / `record_facets`
-
-### Codex
-- 待确认。可能通过 `session_meta.payload.originator` 判断
-- 或者 Codex 不产生独立的 subagent session 文件
-
----
-
-## 8. 实施清单
-
-### Phase 1: Adapter 基础设施（重构，不改行为）
-- [ ] 创建 `scripts/adapters/index.js`
-- [ ] 创建 `scripts/adapters/claude-code.js`（从现有代码提取）
-- [ ] 重构 `scan-sessions.js` 使用 adapter
-- [ ] 重构 `format-session.js` 使用 adapter
-- [ ] 重构 `extract-skills.js` 使用 adapter
-- [ ] 验证：Claude Code 功能不回归
-
-### Phase 2: Codex 支持
-- [ ] 创建 `scripts/adapters/codex-cli.js`
-- [ ] 更新 `postinstall.js` 双平台安装
-- [ ] 更新 `postuninstall.js` 双平台卸载
-- [ ] 用真实 Codex session 文件测试解析
-
-### Phase 3: 文档
-- [ ] 更新 `commands/researchskills-extract.md` 支持动态路径
-- [ ] 更新 `README.md` 双平台说明
-- [ ] 更新 `ResearchSkills/README.md`
diff --git a/researchskills-extract/docs/superpowers/specs/2026-04-13-review-and-score-skills-design.md b/researchskills-extract/docs/superpowers/specs/2026-04-13-review-and-score-skills-design.md
deleted file mode 100644
index b46ae50..0000000
--- a/researchskills-extract/docs/superpowers/specs/2026-04-13-review-and-score-skills-design.md
+++ /dev/null
@@ -1,299 +0,0 @@
-# Design: Review & Score Skills
-
-## Context
-
-`/researchskills-extract` 当前的 pipeline 是 scan → classify → extract (Sonnet) → finalize → upload。Sonnet 提取的 skills 存在三类质量问题：
-
-1. **工程类 skill 泄漏** — 如 Supabase auth deadlock、GitHub Rulesets UI 行为、可视化选型等纯工程内容通过了 tag 过滤
-2. **语义重复** — 名字不同但内容高度重叠的 skills（如 `claim-level-paper-chat-diffing` 和 `claim-level-paper-chat-diffing-for-tacit-knowledge`）
-3. **匿名化不一致** — 部分 skill 的 contributor 字段仍是原始用户名
-
-此外，当前的 `llm_score` 是 Sonnet 自评的单一数字（0-5），无法回答核心问题：**这个 skill 是否真的扩展了最强 AI 的能力边界？**
-
-## 解决方案
-
-在 extract 和 finalize 之间新增两个 Stage，各自独立：
-
-```
-Stage 1 — Scan
-Stage 2 — Classify
-Stage 3 — Extract (Sonnet)
-Stage 4 — Clean (Opus)        ← NEW
-Stage 5 — Score (Opus)        ← NEW
-Stage 6 — Finalize
-Stage 7 — Terminal Summary
-```
-
-- Stage 4 和 Stage 5 各自是一个独立脚本（`clean-skills.js` 和 `score-skills.js`）
-- 各自调用一次 `claude -p --model opus`，Opus 直接操作磁盘文件
-- 两个 Stage 完全隔离：Stage 5 只看到 Stage 4 清洗后存活的文件
-
----
-
-## Stage 4 — Clean Skills
-
-### 脚本
-
-`scripts/clean-skills.js`，安装到 `~/.claude/utils/clean-skills.js`
-
-### 调用方式
-
-```bash
-node ~/.claude/utils/clean-skills.js --session-ids <csv> [--verbose]
-```
-
-### 行为
-
-1. 收集 `~/.researchskills/cache/skills/` 下所有目标 session 的 `.md` 文件
-2. 将所有 skill 内容内联到 prompt 中（每个 skill 用 `<skill path="...">...</skill>` 标签包裹）
-3. 通过 stdin 传给 `claude -p --model opus`
-4. Opus 输出结构化 JSON 决策清单，脚本解析后执行文件操作
-
-> **为什么不是 Opus 直接操作文件？** `claude -p`（pipe mode）是纯文本 I/O，不支持工具调用。所以 Opus 输出决策，脚本执行操作。但对于合并操作，Opus 直接在输出中给出合并后的 skill 全文，脚本写入文件——这样创造性工作仍由 Opus 完成。
-
-Opus 逐个审查每个 skill，输出三类决策：
-
-#### 4a. 剔除不合格的 skills
-
-删除不符合"科学研究知识"定义的 skill 文件。判断标准：
-
-**删除**（这些不是 research skill）：
-- GitHub 平台操作（权限、Rulesets、repo transfer、CI/CD）
-- 前后端工程（auth、数据库、UI 布局、CSS、可视化选型）
-- DevOps / 部署 / 包管理
-- 项目管理 / 术语命名 / 文档组织
-
-**保留**（这些是 research skill）：
-- 科学研究的方法论决策（实验设计、假设选择、数据解释策略）
-- AI/LLM 不知道的领域事实（前沿知识、非公开信息、LLM 常见错误信念的纠偏）
-- 具体的研究转折点（假设被推翻、方法论放弃、意外发现改变方向）
-- 知识表征 / 知识提取的方法论设计（如 schema 设计用于捕获研究决策树）— 当且仅当它讨论的是"如何表征科学知识"而非"如何写代码实现"
-
-#### 4b. 修复匿名化
-
-检查每个存活 skill 的 `contributor` 字段。如果不是 `anon-*` 格式，替换为 `anon-<sha256(原值)前8位>`。
-
-同时扫描 body 内容中可能残留的 PII：
-- 原始用户名
-- 私有 URL（非 arxiv/doi/github/wikipedia/researchskills.ai）
-- 邮箱地址
-- 绝对文件路径
-
-发现则移除或替换为通用描述。
-
-#### 4c. 合并重复 skills
-
-如果两个 skill 的核心内容高度重叠（同一个知识点的不同表述），保留内容更丰富、质量更高的那个，删除另一个。
-
-合并判断依据：
-- 核心 claim / decision / episode 是否本质相同
-- 不同视角描述同一个设计决策 → 合并
-- 相似但有实质不同的 scope 或 context → 不合并
-
-### Opus 输出格式
-
-Opus 在回复末尾输出一个 ````clean-result` fenced JSON block：
-
-```json
-{
-  "actions": [
-    {"action": "keep", "file": "skill-xxx-1.md"},
-    {"action": "reject", "file": "skill-xxx-2.md", "reason": "engineering: Supabase auth debugging"},
-    {"action": "fix_pii", "file": "skill-xxx-3.md", "contributor_hash": "anon-b34ee36c"},
-    {"action": "merge", "keep_file": "skill-xxx-4.md", "delete_file": "skill-xxx-5.md", "merged_content": "---\nname: ...\n---\n(full merged skill text)"}
-  ]
-}
-```
-
-脚本解析 JSON 后执行文件操作：
-- `reject` → 删除文件
-- `fix_pii` → 用正则替换 contributor 字段
-- `merge` → 将 `merged_content` 写入 `keep_file`，删除 `delete_file`
-- `keep` → 不操作
-
-然后打印 human-readable summary。
-
----
-
-## Stage 5 — Score Skills
-
-### 脚本
-
-`scripts/score-skills.js`，安装到 `~/.claude/utils/score-skills.js`
-
-### 调用方式
-
-```bash
-node ~/.claude/utils/score-skills.js --session-ids <csv> [--verbose]
-```
-
-### 行为
-
-1. 收集 Stage 4 存活的 `.md` 文件
-2. 将所有 skill 内容内联到 prompt 中，通过 stdin 传给 `claude -p --model opus`
-3. Opus 输出每个 skill 的 3 维分数（JSON 格式），脚本解析后写入对应文件的 YAML frontmatter
-
-### 三个评分维度
-
-每个维度 0-5 分。写入 YAML frontmatter 的 `review_scores` 字段：
-
-```yaml
-review_scores:
-  procedural: 4
-  semantic: 2
-  episodic: 0
-```
-
-#### procedural — 程序性记忆价值
-
-**问题**：这个 skill 是否提供了最强 AI 不熟悉的研究决策流程或原则？
-
-| 分数 | 含义 |
-|------|------|
-| 0 | AI 完全能自己推导出这个决策逻辑，无需外部提示 |
-| 1 | AI 大概率能想到，但可能不会优先选择这条路径 |
-| 2 | AI 能想到其中一部分，但会遗漏关键的排除条件或失败处理 |
-| 3 | AI 不太可能自主产生这个决策框架，但看到后会认为合理 |
-| 4 | AI 在面对同样的研究 impasse 时会走错路，这个 skill 能直接纠正搜索方向 |
-| 5 | AI 不仅会走错，而且会对错误方向很有信心；这个 skill 纠正的是一个 confident but wrong 的决策模式 |
-
-**关注点**：决策的 trigger condition 是否具体？是否明确拒绝了某些备选方案并给出理由？失败后怎么恢复？
-
-#### semantic — 陈述性记忆价值
-
-**问题**：这个 skill 是否提供了最强 AI 不知道的知识或信念？
-
-| 分数 | 含义 |
-|------|------|
-| 0 | 教科书级知识，任何 LLM 都知道 |
-| 1 | 公开但冷门的知识，模型可能知道但不确定 |
-| 2 | 较新的知识，模型可能在训练截止后才出现 |
-| 3 | 模型大概率不知道的具体事实（如某工具的未文档化行为） |
-| 4 | 模型在此处持有错误信念，会给出自信但错误的回答 |
-| 5 | 非公开的、实验室级别的知识，训练数据中不可能存在 |
-
-**关注点**：这个事实是否有具体的 evidence？是 frontier / non-public / correction 中的哪一类？
-
-#### episodic — 情景性记忆价值
-
-**问题**：这个 skill 是否提供了 AI 可供参考的具体研究经验？
-
-| 分数 | 含义 |
-|------|------|
-| 0 | 纯抽象建议，没有具体情境 |
-| 1 | 有情境描述但很笼统（"在某次实验中……"） |
-| 2 | 有具体的情境和行动，但结果/教训不够明确 |
-| 3 | 完整的情境-行动-结果，AI 在类似情境下可参考 |
-| 4 | 包含反直觉的转折（预期 A 却观察到 B），AI 遇到类似异常时可直接复用 |
-| 5 | 高度具体的失败/适应/异常经历，包含明确的 retrieval cue，AI 在相似情境下能被自动触发 |
-
-**关注点**：是否有具体的 situation → action → outcome 链条？教训是否可迁移？
-
-### 关键设计原则
-
-- **三维独立**：一个 procedural skill 可能同时有高 semantic_value（携带 AI 不知道的事实）和高 episodic_value（来自具体研究 episode）
-- **核心判断标准**：最强的 AI（Opus 级别）在没有这个 skill 的情况下，是否会在对应维度上表现更差？
-- **不是 memory_type 的重复**：`memory_type` 是 skill 的结构分类（它的格式像什么），`review_scores` 是内容价值评估（它真的带来了什么）
-
-### 输出
-
-操作完毕后，Opus 输出 summary：
-
-```
-SCORE_RESULT: {"scored": 12, "scores": [{"name": "...", "procedural": 4, "semantic": 2, "episodic": 0}, ...]}
-```
-
----
-
-## 对 researchskills-extract.md 命令文件的修改
-
-### Pipeline 更新
-
-```
-scan-sessions.js       ─┐
-classify-projects.js   ─┤
-extract-skills.js      ─┤  deterministic scripts (you call them)
-clean-skills.js        ─┤  ← NEW: Opus cleans extracted skills
-score-skills.js        ─┤  ← NEW: Opus scores surviving skills
-finalize.js            ─┘
-```
-
-### Helper scripts 表格新增
-
-| Script | What it does |
-|--------|-------------|
-| `clean-skills.js` | Review extracted skills with Opus: reject engineering, fix PII, merge duplicates |
-| `score-skills.js` | Score surviving skills with Opus on 3 dimensions: procedural, semantic, episodic value |
-
-### 新增 Stage 4 — Clean Skills
-
-```bash
-node ~/.claude/utils/clean-skills.js \
-  --session-ids <ALL-research-session-ids-csv> \
-  --verbose
-```
-
-读取输出，报告：`"Cleaned: kept N, rejected M, merged K."`
-
-### 新增 Stage 5 — Score Skills
-
-```bash
-node ~/.claude/utils/score-skills.js \
-  --session-ids <ALL-research-session-ids-csv> \
-  --verbose
-```
-
-读取输出，报告：`"Scored N skills. Average: procedural X.X, semantic X.X, episodic X.X."`
-
-### Stage 编号调整
-
-- Stage 1 — Scan（不变）
-- Stage 2 — Classify（不变）
-- Stage 3 — Extract（不变）
-- Stage 4 — Clean Skills（新）
-- Stage 5 — Score Skills（新）
-- Stage 6 — Finalize（原 Stage 4）
-- Stage 7 — Terminal Summary（原 Stage 5，新增 review_scores 统计）
-
-### Terminal Summary 更新
-
-```
-═══════════════════════════════════════════════════════
-  /researchskills-extract Complete!
-═══════════════════════════════════════════════════════
-
-Extracted N skills from M sessions across P projects:
-  • Episodic:   E skills
-  • Semantic:   S skills
-  • Procedural: Pr skills
-
-Review (Opus):
-  • Kept: K / Rejected: R / Merged: G
-  • Avg scores: procedural X.X, semantic X.X, episodic X.X
-
-Review your skills:
-  → https://researchskills.ai/review/batch/<batchId>
-═══════════════════════════════════════════════════════
-```
-
----
-
-## 文件清单
-
-### 新建
-- `scripts/clean-skills.js` — Stage 4 脚本
-- `scripts/score-skills.js` — Stage 5 脚本
-
-### 修改
-- `commands/researchskills-extract.md` — 插入 Stage 4、5，调整后续编号
-- `scripts/postinstall.js` — HELPER_SCRIPTS 数组新增两个脚本
-- `scripts/postuninstall.js` — 清理新增脚本（如有清理逻辑）
-
----
-
-## 验证
-
-1. 在现有 cache 上运行 `clean-skills.js`，确认工程类 skills 被删除、PII 被修复、重复被合并
-2. 在清洗后的 cache 上运行 `score-skills.js`，确认每个 skill 的 YAML 中有 `review_scores` 字段
-3. 运行完整 `/researchskills-extract` 流程，确认 7 个 Stage 顺序正确，Terminal Summary 包含 review 统计
-4. 检查上传到 researchskills.ai 的 skills 是否携带 `review_scores`
diff --git a/researchskills-extract/package.json b/researchskills-extract/package.json
deleted file mode 100644
index 5f8552e..0000000
--- a/researchskills-extract/package.json
+++ /dev/null
@@ -1,41 +0,0 @@
-{
-  "name": "@scienceintelligence/researchskills-extract",
-  "version": "0.18.1",
-  "description": "Extract your research trajectory as a decision tree from conversation history for ResearchSkills. Supports both Claude Code and Codex.",
-  "keywords": [
-    "claude-code",
-    "codex",
-    "researchskills",
-    "scientific-research",
-    "know-how",
-    "skills",
-    "ai-science"
-  ],
-  "author": "ResearchSkills Contributors",
-  "license": "CC-BY-4.0",
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/ScienceIntelligence/ResearchSkills.git",
-    "directory": "researchskills-extract"
-  },
-  "publishConfig": {
-    "access": "public"
-  },
-  "scripts": {
-    "postinstall": "node scripts/postinstall.js",
-    "postuninstall": "node scripts/postuninstall.js",
-    "test": "node tests/test-postinstall.js"
-  },
-  "files": [
-    "commands/",
-    "templates/",
-    "scripts/",
-    "tests/"
-  ],
-  "engines": {
-    "node": ">=18.0.0"
-  },
-  "dependencies": {
-    "@anthropic-ai/sdk": "^0.88.0"
-  }
-}
diff --git a/researchskills-extract/scripts/classify-projects.js b/researchskills-extract/scripts/classify-projects.js
deleted file mode 100644
index cf78bdb..0000000
--- a/researchskills-extract/scripts/classify-projects.js
+++ /dev/null
@@ -1,284 +0,0 @@
-#!/usr/bin/env node
-/**
- * classify-projects.js
- *
- * Classifies projects from work-list.json as research vs engineering
- * using AI (via platform.js), then picks domain/subdomain from the taxonomy.
- *
- * Usage:
- *   classify-projects.js <work-list.json> [--test] [--verbose]
- *
- * Output: ~/.researchskills/cache/classification.json
- *   {
- *     projects: {
- *       "<project_path>": {
- *         slug: "project-name",
- *         type: "research" | "engineering" | "other",
- *         domain: "computer-science",
- *         subdomain: "artificial-intelligence",
- *         session_ids: ["id1", "id2", ...],
- *         research_session_ids: ["id1", ...],
- *         skipped_session_ids: ["id3", ...],
- *         reason: "why classified this way"
- *       }
- *     }
- *   }
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const https = require('https');
-const { parsePlatformFlag, createRunner } = require('./platform');
-
-const OUTPUT_PATH = path.join(os.homedir(), '.researchskills', 'cache', 'classification.json');
-
-// Max concurrent AI classify calls — prevents resource thrashing & API rate limits
-const CONCURRENCY = 5;
-
-// ---------------------------------------------------------------------------
-// CLI
-// ---------------------------------------------------------------------------
-
-function parseArgs() {
-  const args = process.argv.slice(2);
-  const opts = { workListPath: null, test: false, verbose: false };
-  for (let i = 0; i < args.length; i++) {
-    switch (args[i]) {
-      case '--test':    opts.test = true; break;
-      case '--verbose': opts.verbose = true; break;
-      default:
-        if (!args[i].startsWith('-') && !opts.workListPath) {
-          opts.workListPath = args[i];
-        }
-    }
-  }
-  return opts;
-}
-
-// ---------------------------------------------------------------------------
-// Fetch taxonomy
-// ---------------------------------------------------------------------------
-
-function fetchTaxonomy() {
-  return new Promise((resolve, reject) => {
-    https.get('https://researchskills.ai/taxonomy.json', (res) => {
-      let data = '';
-      res.on('data', (c) => data += c);
-      res.on('end', () => {
-        try {
-          resolve(JSON.parse(data).taxonomy);
-        } catch (e) {
-          reject(new Error('Failed to parse taxonomy: ' + e.message));
-        }
-      });
-    }).on('error', reject);
-  });
-}
-
-// ---------------------------------------------------------------------------
-// Build classification prompt for one project
-// ---------------------------------------------------------------------------
-
-function buildPrompt(projectPath, sessions, taxonomyStr, isTest) {
-  const slug = projectPath.split('/').filter(Boolean).pop() || 'unknown';
-
-  // Gather prompt samples: first_prompt + sampled_prompts from up to 5 sessions
-  const samples = [];
-  const picked = sessions.slice(0, 5);
-  for (const s of picked) {
-    const lines = [`[Session ${s.session_id.substring(0, 8)} | ${s.user_message_count} msgs | ${Math.round(s.duration_minutes)}min]`];
-    if (s.first_prompt) lines.push(`  First: ${String(s.first_prompt).substring(0, 300)}`);
-    const sp = s.sampled_prompts || [];
-    for (const p of sp.slice(0, 2)) {
-      lines.push(`  Sample: ${String(p).substring(0, 200)}`);
-    }
-    samples.push(lines.join('\n'));
-  }
-
-  return `Classify this project and pick the best domain/subdomain.
-
-## Project
-Path slug: ${slug}
-Sessions: ${sessions.length} total
-
-## Message samples (from up to 5 sessions)
-${samples.join('\n\n')}
-
-## Available domains/subdomains
-${taxonomyStr}
-
-## Task
-${isTest
-    ? 'TEST MODE: Accept both research AND engineering projects. Map engineering to computer-science/software-engineering or the closest match.'
-    : 'PRODUCTION MODE: Only classify as "research" if the sessions involve genuine scientific inquiry, research methodology, hypothesis testing, academic writing, OR computational science where the code is written to answer or validate a research question (simulations, solvers, numerical methods, scientific data analysis pipelines). Reusable numerics libraries or benchmark harnesses with no specific scientific inquiry are engineering, not research. Software engineering (web dev, deployment, debugging non-scientific tools, UI work, auth, package management) is NOT research regardless of difficulty.'}
-
-Respond with EXACTLY this JSON (no markdown fences, no other text):
-{"type":"research","domain":"...","subdomain":"...","project_name":"...","reason":"one sentence why","skip_patterns":["pattern1"]}
-
-- type: "research" or "engineering" or "other"
-- domain/subdomain: from the taxonomy list above. For engineering use "computer-science/software-engineering"
-- project_name: a short, descriptive name (3-8 words) summarizing the research topic of this project based on the session content. Do NOT use the folder name. Examples: "Protein Folding Simulation Pipeline", "Neural ODE Parameter Estimation", "Galaxy Merger Classification". For engineering projects, describe the tool/system being built.
-- reason: one sentence explaining classification
-- skip_patterns: substrings in first_prompt that indicate non-research sessions to skip (e.g. "researchskills-extract", "extract-knowhow", "npm run build"). Empty array if none.`;
-}
-
-// ---------------------------------------------------------------------------
-// Main
-// ---------------------------------------------------------------------------
-
-async function main() {
-  const platform = parsePlatformFlag();
-  const runner = createRunner(platform);
-  const opts = parseArgs();
-  if (!opts.workListPath) {
-    console.error('Usage: classify-projects.js <work-list.json> [--cc|--codex] [--test] [--verbose]');
-    process.exit(1);
-  }
-
-  const workList = JSON.parse(fs.readFileSync(opts.workListPath, 'utf-8'));
-  const sessions = workList.sessions || [];
-  const projectMap = workList.projects || {};
-
-  // Fetch taxonomy
-  let taxonomy;
-  try {
-    taxonomy = await fetchTaxonomy();
-  } catch (e) {
-    console.error(`⚠ Failed to fetch taxonomy: ${e.message}. Using fallback.`);
-    taxonomy = { 'computer-science': ['artificial-intelligence', 'software-engineering'] };
-  }
-  const taxonomyStr = Object.entries(taxonomy)
-    .sort(([a], [b]) => a.localeCompare(b))
-    .map(([d, subs]) => `${d}: ${subs.join(', ')}`)
-    .join('\n');
-
-  // Group sessions by project
-  const byProject = {};
-  for (const s of sessions) {
-    const p = s.project_path;
-    if (!byProject[p]) byProject[p] = [];
-    byProject[p].push(s);
-  }
-
-  const result = { projects: {} };
-  const projectPaths = Object.keys(byProject);
-
-  console.log(`\nClassifying ${projectPaths.length} projects (${CONCURRENCY} at a time)...\n`);
-
-  // Classify projects with bounded concurrency
-  const classifyOne = async (projPath) => {
-    const projSessions = byProject[projPath];
-    const slug = projPath.split('/').filter(Boolean).pop() || 'unknown';
-
-    const prompt = buildPrompt(projPath, projSessions, taxonomyStr, opts.test);
-    const { ok, output, error } = await runner.classify(prompt);
-
-    if (!ok) {
-      return {
-        projPath, slug, type: 'error', domain: null, subdomain: null,
-        session_ids: projSessions.map(s => s.session_id),
-        research_session_ids: [],
-        skipped_session_ids: projSessions.map(s => s.session_id),
-        reason: `Classification failed: ${error}`,
-      };
-    }
-
-    // Parse JSON from Sonnet output.
-    // Use bracket-balanced extraction instead of a greedy regex — if any text
-    // follows the JSON (e.g. a Claude Code hook prints to stdout after the
-    // response), the greedy /\{[\s\S]*\}/ would stretch to the last "}" in
-    // that trailing text and produce invalid JSON for JSON.parse.
-    let classification;
-    try {
-      const start = output.indexOf('{');
-      if (start === -1) throw new Error('No JSON found in output');
-      let depth = 0, end = -1;
-      for (let i = start; i < output.length; i++) {
-        if (output[i] === '{') depth++;
-        else if (output[i] === '}') { depth--; if (depth === 0) { end = i; break; } }
-      }
-      if (end === -1) throw new Error('Unbalanced JSON object in output');
-      classification = JSON.parse(output.slice(start, end + 1));
-    } catch (e) {
-      if (opts.verbose) console.log(`  ${slug}: PARSE ERROR — ${e.message}\n    Output: ${output.substring(0, 200)}`);
-      return {
-        projPath, slug, type: 'error', domain: null, subdomain: null,
-        session_ids: projSessions.map(s => s.session_id),
-        research_session_ids: [],
-        skipped_session_ids: projSessions.map(s => s.session_id),
-        reason: `Parse error: ${e.message}`,
-      };
-    }
-
-    // Filter sessions using skip_patterns
-    const skipPatterns = classification.skip_patterns || [];
-    const researchIds = [];
-    const skippedIds = [];
-    for (const s of projSessions) {
-      const fp = String(s.first_prompt || '');
-      if (skipPatterns.some(pat => fp.includes(pat))) {
-        skippedIds.push(s.session_id);
-      } else {
-        researchIds.push(s.session_id);
-      }
-    }
-
-    return {
-      projPath, slug,
-      type: classification.type || 'other',
-      domain: classification.domain || null,
-      subdomain: classification.subdomain || null,
-      project_name: classification.project_name || null,
-      session_ids: projSessions.map(s => s.session_id),
-      research_session_ids: researchIds,
-      skipped_session_ids: skippedIds,
-      reason: classification.reason || '',
-    };
-  };
-
-  // Process in batches of CONCURRENCY to avoid spawning too many CLI processes
-  const classifications = [];
-  for (let i = 0; i < projectPaths.length; i += CONCURRENCY) {
-    const batch = projectPaths.slice(i, i + CONCURRENCY);
-    const batchResults = await Promise.all(batch.map(classifyOne));
-    classifications.push(...batchResults);
-    if (i + CONCURRENCY < projectPaths.length) {
-      console.log(`  ... classified ${Math.min(i + CONCURRENCY, projectPaths.length)}/${projectPaths.length} projects`);
-    }
-  }
-
-  // Collect results and print
-  for (const c of classifications) {
-    result.projects[c.projPath] = c;
-
-    const tag = c.type === 'research' ? '✓ RESEARCH' :
-                c.type === 'engineering' ? '✗ engineering' :
-                c.type === 'error' ? '✗ ERROR' : '? other';
-    console.log(`  ${c.slug} (${c.session_ids.length} sessions): ${tag}`);
-    if (c.project_name) console.log(`    → "${c.project_name}"`);
-    if (c.domain) console.log(`    → ${c.domain}/${c.subdomain}`);
-    if (c.skipped_session_ids.length > 0) {
-      console.log(`    → ${c.research_session_ids.length} research, ${c.skipped_session_ids.length} skipped`);
-    }
-    if (opts.verbose) console.log(`    Reason: ${c.reason}`);
-  }
-
-  // Write output
-  fs.writeFileSync(OUTPUT_PATH, JSON.stringify(result, null, 2));
-  console.log(`\n✓ Classification saved to ${OUTPUT_PATH}`);
-
-  // Summary
-  const projects = Object.values(result.projects);
-  const research = projects.filter(p => p.type === 'research');
-  const totalResearchSessions = research.reduce((n, p) => n + p.research_session_ids.length, 0);
-  console.log(`\n  ${research.length}/${projects.length} projects classified as research`);
-  console.log(`  ${totalResearchSessions} research sessions to extract\n`);
-}
-
-main().catch(err => {
-  console.error(`Fatal: ${err.message}`);
-  process.exit(1);
-});
diff --git a/researchskills-extract/scripts/clean-skills.js b/researchskills-extract/scripts/clean-skills.js
deleted file mode 100644
index 452632a..0000000
--- a/researchskills-extract/scripts/clean-skills.js
+++ /dev/null
@@ -1,202 +0,0 @@
-#!/usr/bin/env node
-/**
- * clean-skills.js
- *
- * Stage 4 of /researchskills-extract: review extracted skills.
- * Spawns an AI instance (via platform.js) that directly
- * reads, deletes, edits, and merges skill files on disk.
- *
- * Operations:
- *   - Reject engineering skills (delete files)
- *   - Fix PII / anonymization (edit files)
- *   - Merge duplicate skills (write merged, delete redundant)
- *
- * Usage:
- *   clean-skills.js --session-ids id1,id2,... [--verbose]
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const { parsePlatformFlag, createRunner } = require('./platform');
-
-const CACHE_DIR = path.join(os.homedir(), '.researchskills', 'cache', 'skills');
-
-// ---------------------------------------------------------------------------
-// CLI
-// ---------------------------------------------------------------------------
-
-function parseArgs() {
-  const args = process.argv.slice(2);
-  const opts = { sessionIds: null, verbose: false };
-  for (let i = 0; i < args.length; i++) {
-    switch (args[i]) {
-      case '--session-ids': opts.sessionIds = args[++i].split(','); break;
-      case '--verbose':     opts.verbose = true; break;
-    }
-  }
-  return opts;
-}
-
-// ---------------------------------------------------------------------------
-// Collect skill files
-// ---------------------------------------------------------------------------
-
-function collectSkillFiles(sessionIds) {
-  const files = [];
-  for (const sid of sessionIds) {
-    const dir = path.join(CACHE_DIR, sid);
-    if (!fs.existsSync(dir)) continue;
-    for (const f of fs.readdirSync(dir)) {
-      if (f.endsWith('.md')) {
-        files.push(path.join(dir, f));
-      }
-    }
-  }
-  return files;
-}
-
-// ---------------------------------------------------------------------------
-// Build prompt
-// ---------------------------------------------------------------------------
-
-function buildPrompt(files) {
-  const fileList = files.map(f => f).join('\n');
-
-  return `You are a research knowledge quality reviewer. You have ${files.length} skill files to review.
-
-These files were auto-extracted from scientist-AI conversations by a weaker model. Your job is to clean them up: reject non-research content, fix PII leaks, and merge duplicates.
-
-## Skill file paths
-
-<files>
-${fileList}
-</files>
-
-## Instructions
-
-Read each file using the Read tool, then perform these operations:
-
-### 1. Reject non-research skills
-
-DELETE (using Bash: rm <filepath>) any skill that is fundamentally engineering, not research:
-- GitHub platform operations (permissions, Rulesets, repo transfer, CI/CD, branch protection)
-- Frontend/backend engineering (auth, database, UI layout, CSS, visualization library choices)
-- DevOps / deployment / package management
-- Project management / terminology naming / documentation organization
-- Debugging specific software tools (Supabase, Firebase, React, etc.)
-
-KEEP skills about:
-- Scientific research methodology decisions (experiment design, hypothesis selection, data interpretation)
-- Domain facts that AI doesn't know (frontier knowledge, unpublished info, corrections to LLM misconceptions)
-- Concrete research turning points (hypothesis overturned, methodology abandoned, unexpected findings)
-- Knowledge representation methodology (schema design for capturing research decision trees) — ONLY when discussing "how to represent scientific knowledge", NOT "how to code the implementation"
-- Computational science methods (numerical algorithms, solver configurations, simulation setup, scientific data analysis and preprocessing) — where the code implements or validates a scientific method, not generic software engineering
-
-### 2. Check for residual PII
-
-The \`contributor\` field in YAML frontmatter should be the contributor's GitHub handle (real identity is expected here — it is stored separately in the DB column, not in the skill body).
-
-Scan the **body** for residual PII:
-- Real usernames or person names
-- Private URLs (not arxiv.org, doi.org, github.com, en.wikipedia.org, researchskills.ai)
-- Email addresses
-- Absolute file paths (e.g., /Users/...)
-
-If found, use Edit to remove or replace with generic descriptions.
-
-### 3. Merge duplicates
-
-If two skills cover the same core knowledge point (same claim/decision/episode, just different wording or perspective):
-- Keep the one with richer content and higher quality
-- Use Write to save the merged/improved version to the kept file path
-- Delete the redundant file with Bash rm
-
-Do NOT merge skills that have genuinely different scope or context, even if they share some keywords.
-
-### 4. Output summary
-
-After completing ALL operations, output this exact line as the very last line of your response:
-
-CLEAN_SUMMARY: kept=<number> rejected=<number> merged=<number> pii_fixed=<number>
-
-Where:
-- kept = files that survived without changes (or only PII fixes)
-- rejected = files deleted as non-research
-- merged = files deleted because they were merged into another
-- pii_fixed = files where PII was corrected
-
-Now begin. Read each file and process them systematically.`;
-}
-
-// ---------------------------------------------------------------------------
-// Parse summary from output
-// ---------------------------------------------------------------------------
-
-function parseSummary(output) {
-  const match = output.match(/CLEAN_SUMMARY:\s*kept=(\d+)\s+rejected=(\d+)\s+merged=(\d+)\s+pii_fixed=(\d+)/);
-  if (!match) return null;
-  return {
-    kept: parseInt(match[1], 10),
-    rejected: parseInt(match[2], 10),
-    merged: parseInt(match[3], 10),
-    pii_fixed: parseInt(match[4], 10),
-  };
-}
-
-// ---------------------------------------------------------------------------
-// Main
-// ---------------------------------------------------------------------------
-
-async function main() {
-  const platform = parsePlatformFlag();
-  const runner = createRunner(platform);
-  const opts = parseArgs();
-
-  if (!opts.sessionIds || opts.sessionIds.length === 0) {
-    console.error('Usage: clean-skills.js --session-ids id1,id2,... [--cc|--codex] [--verbose]');
-    process.exit(1);
-  }
-
-  const files = collectSkillFiles(opts.sessionIds);
-  console.log(`\nClean: ${files.length} skill files across ${opts.sessionIds.length} sessions`);
-
-  if (files.length === 0) {
-    console.log('No skill files to clean. Done.');
-    process.exit(0);
-  }
-
-  console.log('Spawning AI for review...\n');
-  const prompt = buildPrompt(files);
-  const { ok, output, error } = await runner.clean(prompt, opts.verbose);
-
-  if (!ok) {
-    console.error(`\nReview failed: ${error}`);
-    process.exit(1);
-  }
-
-  const summary = parseSummary(output);
-  if (summary) {
-    console.log(`
-═══════════════════════════════════════════════
-  Clean Complete
-═══════════════════════════════════════════════
-  Kept:      ${summary.kept}
-  Rejected:  ${summary.rejected}
-  Merged:    ${summary.merged}
-  PII fixed: ${summary.pii_fixed}
-═══════════════════════════════════════════════`);
-  } else {
-    console.log('\nCompleted but no CLEAN_SUMMARY found in output.');
-    if (!opts.verbose) {
-      console.log('Re-run with --verbose to see full output.');
-    }
-  }
-}
-
-main().catch(err => {
-  console.error('Fatal:', err);
-  process.exit(1);
-});
diff --git a/researchskills-extract/scripts/extract-skills.js b/researchskills-extract/scripts/extract-skills.js
deleted file mode 100644
index d1760a5..0000000
--- a/researchskills-extract/scripts/extract-skills.js
+++ /dev/null
@@ -1,586 +0,0 @@
-#!/usr/bin/env node
-/**
- * extract-skills.js
- *
- * Orchestrates per-session skill extraction using AI CLI (via platform.js).
- * For each uncached session: format → call AI → validate.
- *
- * This script is the deterministic loop; Sonnet does the AI judgment.
- * The main /researchskills-extract agent calls this script once with Bash,
- * keeping its own context clean.
- *
- * Usage:
- *   extract-skills.js <work-list.json> [options]
- *
- * Options:
- *   --domain <domain>          Domain for all skills (e.g. computer-science)
- *   --subdomain <subdomain>    Subdomain for all skills (e.g. machine-learning)
- *   --contributor <name>       Contributor name (git user.name)
- *   --project-name <name>     Project name (default: derived from session project_path)
- *   --project-slug <slug>     Project slug (default: derived from project-name)
- *   --batch-size <n>           Max sessions to process (default: unlimited)
- *   --concurrency <n>          Parallel claude -p calls per batch (default: 5)
- *   --session-ids <csv>        Only process these session IDs (comma-separated)
- *   --test                     Mark as test data
- *   --verbose                  Print detailed progress
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const { execFileSync } = require('child_process');
-const { parsePlatformFlag, createRunner } = require('./platform');
-
-const FORMAT_SCRIPT = path.join(__dirname, 'format-session.js');
-const VALIDATE_SCRIPT = path.join(__dirname, 'validate-skills.js');
-
-// Pre-filter: skip sessions too short to contain meaningful research content
-const MIN_FORMATTED_CHARS = 2000;
-
-// Skip sessions whose first_prompt matches these patterns (case-insensitive)
-const SKIP_PROMPT_PATTERNS = [
-  /^(hi|hello|hey|test|help)\s*[.!?]?\s*$/i,
-  /^\/\w+/,                           // slash command invocations
-  /^(git|npm|pip|brew|docker)\s/i,    // pure tool commands
-  /^(fix|install|update|upgrade)\s+(the\s+)?(lint|build|ci|deps|dependencies)/i,
-];
-
-// ---------------------------------------------------------------------------
-// CLI argument parsing
-// ---------------------------------------------------------------------------
-
-function parseArgs() {
-  const args = process.argv.slice(2);
-  const opts = {
-    workListPath: null,
-    domain: null,
-    subdomain: null,
-    contributor: null,
-    projectName: null,
-    projectSlug: null,
-    batchSize: Infinity,
-    concurrency: 5,
-    sessionIds: null,
-    singleBatch: false,
-    test: false,
-    verbose: false,
-  };
-
-  for (let i = 0; i < args.length; i++) {
-    switch (args[i]) {
-      case '--domain':        opts.domain = args[++i]; break;
-      case '--subdomain':     opts.subdomain = args[++i]; break;
-      case '--contributor':   opts.contributor = args[++i]; break;
-      case '--project-name':  opts.projectName = args[++i]; break;
-      case '--project-slug':  opts.projectSlug = args[++i]; break;
-      case '--batch-size':    opts.batchSize = parseInt(args[++i], 10); break;
-      case '--concurrency':   opts.concurrency = parseInt(args[++i], 10); break;
-      case '--session-ids':   opts.sessionIds = args[++i].split(','); break;
-      case '--single-batch':  opts.singleBatch = true; break;
-      case '--test':          opts.test = true; break;
-      case '--verbose':       opts.verbose = true; break;
-      default:
-        if (!args[i].startsWith('-') && !opts.workListPath) {
-          opts.workListPath = args[i];
-        }
-    }
-  }
-  return opts;
-}
-
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-
-function run(cmd, args, opts = {}) {
-  try {
-    return execFileSync(cmd, args, {
-      encoding: 'utf-8',
-      stdio: opts.stdio || ['pipe', 'pipe', 'pipe'],
-      timeout: opts.timeout || 300_000, // 5 min default
-      ...opts,
-    }).trim();
-  } catch (err) {
-    if (opts.ignoreError) return err.stdout || '';
-    throw err;
-  }
-}
-
-function isCached(sessionId) {
-  try {
-    execFileSync('node', [VALIDATE_SCRIPT, 'is-cached', sessionId], {
-      stdio: 'pipe',
-    });
-    return true;
-  } catch { return false; }
-}
-
-function isSegmentCached(sessionId, segInfo) {
-  const dir = path.join(os.homedir(), '.researchskills', 'cache', 'skills', sessionId);
-  if (!segInfo) {
-    // Single-segment session: cached if skills exist OR .done marker present
-    try {
-      const entries = fs.readdirSync(dir);
-      if (entries.includes('.done')) return true;
-      return entries.filter(f => f.endsWith('.md')).some(f => !/-s\d+-/.test(f));
-    } catch { return false; }
-  }
-  // Multi-segment: check for segment-specific .done marker or skill files
-  try {
-    const entries = fs.readdirSync(dir);
-    if (entries.includes(`.done-s${segInfo.index}`)) return true;
-    const prefix = `-s${segInfo.index}-`;
-    return entries.filter(f => f.endsWith('.md') && f.includes(prefix)).length > 0;
-  } catch { return false; }
-}
-
-function formatSession(filePath, outputPath) {
-  const stdout = run('node', [FORMAT_SCRIPT, filePath, outputPath]);
-  try { return JSON.parse(stdout); } catch { return null; }
-}
-
-// ---------------------------------------------------------------------------
-// Extraction prompt builder
-// ---------------------------------------------------------------------------
-
-function deriveProjectName(session, opts) {
-  if (opts.projectName) return opts.projectName;
-  // Derive from project_path (the cwd or CC project directory)
-  const pp = session.project_path;
-  if (!pp) return null;
-  // Use the last path component as project name
-  return path.basename(pp);
-}
-
-function slugify(s) {
-  return String(s || '')
-    .toLowerCase()
-    .replace(/[^a-z0-9]+/g, '-')
-    .replace(/^-+|-+$/g, '')
-    .substring(0, 60);
-}
-
-function buildPrompt(segmentFile) {
-  const sessionText = fs.readFileSync(segmentFile, 'utf-8');
-
-  return `Extract research skills from this scientist-AI conversation. Output 0–10 skills. 0 is often correct.
-
-## Quality Gate — ALL three must pass
-
-1. **Non-obvious** — would a domain expert find this surprising?
-2. **Session-specific** — grounded in a concrete event here, not generic advice?
-3. **Research-only** — about scientific methodology, reasoning, knowledge, OR computational science methods (simulations, solvers, numerical algorithms, scientific data analysis)? Reject pure software engineering (UI, DevOps, database, build tools, git, auth, package management) even if the project is research-related. But DO extract skills about scientific computing where the code implements or validates a scientific method.
-
-## Boundary examples — same project, different verdicts
-
-A research project contains BOTH research and engineering work. Extract only the research.
-
-❌ REJECT (pure engineering):
-- "Supabase onAuthStateChange callback deadlocks on re-entry" → auth/infra debugging
-- "GitHub Rulesets vs Branch Protection Rules show different UI warnings" → platform ops
-- "Static PNG unreadable at 130+ nodes, switch to interactive HTML/JS tree" → UI/visualization choice
-- "Repository transfer converts collaborators to Outside Collaborators" → GitHub admin
-- "README tables don't scale beyond 10 contributors" → UI/UX layout
-
-✅ EXTRACT (research):
-- "Claim-level diffing between paper and chat logs reveals unpublished research decisions" → methodology for tacit knowledge capture
-- "LLMs fail at research due to task structure mismatch (long horizon + no verifier), not reasoning deficits" → frontier knowledge about AI capabilities
-- "Predefined 20-action vocabulary with escape hatch balances structure vs flexibility for research ontology" → schema design for knowledge representation
-
-✅ EXTRACT (computational science — code IS the science):
-- "FFT-based spectral method converges 10x faster than finite-difference for periodic MHD boundary conditions" → numerical method selection for domain physics
-- "Helmholtz decomposition requires divergence-free initial field; naive random initialization violates ∇·B=0" → constraint on simulation setup
-- "Tracking magnetic field lines near coil edges fails with RK4 step > 0.01 due to high curvature" → numerical stability boundary for specific geometry
-- "Alfvénic wave flux from PSP data requires detrending at spacecraft spin period before spectral analysis" → data preprocessing for space physics
-
-When in doubt, ask: "Is this about HOW to do science (including computational science), or HOW to build software unrelated to the research question?" Only the former qualifies.
-
-## Skill types
-
-Output each skill as a \`\`\`skill-md fenced block. YAML frontmatter: name, memory_type, subtype, llm_score (0–5), tags. Then markdown sections.
-
-**procedural** — IF-THEN rules for research impasses (Soar impasse taxonomy):
-- \`tie\`: Multiple viable paths, agent cannot rank them. Provide decision heuristics to break the tie.
-- \`no-change\`: Completely stuck, no candidate action. Provide exploration strategies or problem reframing.
-- \`constraint-failure\`: A methodological assumption turns out violated. Provide applicability boundaries and alternatives.
-- \`operator-fail\`: Correct approach selected but execution fails. Provide diagnostics and failure patterns.
-Sections: When + Exclusions → Decision (Preferred / Rejected / Reasoning) → Local Verifiers → Failure Handling → Anti-exemplars
-
-**semantic** — Facts missing from LLM training data (ACT-R declarative memory):
-- \`frontier\`: Post-training-cutoff knowledge — fact does not exist in model weights.
-- \`non-public\`: Lab-internal or unpublished — never in any training corpus.
-- \`correction\`: LLM actively gets this wrong — confident but incorrect default belief.
-Sections: Fact → Evidence → LLM Default Belief (correction only) → Expiry Signal
-
-**episodic** — Concrete research episodes as situation-action-outcome triples (CBR, Kolodner 1993):
-- \`failure\`: Attempted X, failed due to hidden reason Y. Trigger: agent about to do something similar.
-- \`adaptation\`: Standard method failed, workaround Z succeeded. Trigger: agent stuck with standard approach.
-- \`anomalous\`: Expected A, observed B, turned out important. Trigger: agent observes similar anomaly.
-Sections: Situation → Action → Outcome → Lesson → Retrieval Cues
-
-## Rules
-
-- De-identify: strip paths, usernames, project names, private URLs. Paraphrase non-English quotes in English. Preserve scientific content (materials, parameters, methods, public libraries).
-- llm_score: 4–5 genuinely novel, 3 useful, 0–2 skip.
-
-## Final line
-
-SKILLS_EXTRACTED: <total> (E:<episodic> S:<semantic> P:<procedural>)
-
-<session>
-${sessionText}
-</session>`;
-}
-
-// ---------------------------------------------------------------------------
-// Parse skill-md blocks from Sonnet output → write files → validate
-// ---------------------------------------------------------------------------
-
-function injectProjectMeta(content, projectMeta) {
-  if (!projectMeta || !content.startsWith('---')) return content;
-  const lines = content.split('\n');
-  const endIdx = lines.indexOf('---', 1);
-  if (endIdx <= 0) return content;
-  const insert = [];
-  // Inject fixed fields that Sonnet no longer outputs
-  if (projectMeta.domain) insert.push(`domain: ${projectMeta.domain}`);
-  if (projectMeta.subdomain) insert.push(`subdomain: ${projectMeta.subdomain}`);
-  if (projectMeta.contributor) insert.push(`contributor: ${projectMeta.contributor}`);
-  if (projectMeta.sessionId) {
-    insert.push(`source:`);
-    insert.push(`  type: session`);
-    insert.push(`  session_id: "${projectMeta.sessionId}"`);
-  }
-  if (projectMeta.extractedAt) insert.push(`extracted_at: "${projectMeta.extractedAt}"`);
-  if (projectMeta.projectName) insert.push(`project_name: "${projectMeta.projectName}"`);
-  if (projectMeta.projectSlug) insert.push(`project_slug: "${projectMeta.projectSlug}"`);
-  if (insert.length === 0) return content;
-  lines.splice(endIdx, 0, ...insert);
-  return lines.join('\n');
-}
-
-function markSegmentDone(sessionId, segInfo) {
-  const dir = path.join(os.homedir(), '.researchskills', 'cache', 'skills', sessionId);
-  fs.mkdirSync(dir, { recursive: true });
-  const marker = segInfo ? `.done-s${segInfo.index}` : '.done';
-  fs.writeFileSync(path.join(dir, marker), '');
-}
-
-function writeAndValidateSkills(sessionId, output, projectMeta, segInfo) {
-  const blocks = [];
-  const regex = /```skill-md\n([\s\S]*?)```/g;
-  let m;
-  while ((m = regex.exec(output)) !== null) {
-    blocks.push(m[1].trim());
-  }
-
-  if (blocks.length === 0) {
-    markSegmentDone(sessionId, segInfo);
-    return 0;
-  }
-
-  const segPrefix = segInfo ? `s${segInfo.index}-` : '';
-  const skillFiles = [];
-  for (let i = 0; i < blocks.length; i++) {
-    const content = injectProjectMeta(blocks[i], projectMeta);
-    const filePath = path.join(os.tmpdir(), `skill-${sessionId}-${segPrefix}${i + 1}.md`);
-    fs.writeFileSync(filePath, content + '\n');
-    skillFiles.push(filePath);
-  }
-
-  // Validate and cache via execFileSync (no shell)
-  try {
-    execFileSync('node', [VALIDATE_SCRIPT, 'save', sessionId, ...skillFiles], {
-      encoding: 'utf-8',
-      stdio: 'pipe',
-    });
-  } catch {
-    // Validation failed — skills still written but not cached
-  }
-
-  // Clean up tmp files
-  for (const f of skillFiles) { try { fs.unlinkSync(f); } catch {} }
-
-  return blocks.length;
-}
-
-// ---------------------------------------------------------------------------
-// Process a single session: pre-filter → format → build prompt
-// Returns { skip, error, ready, sid, prompt, formattedFiles }
-// ---------------------------------------------------------------------------
-
-function prepareSession(session) {
-  const sid = session.session_id;
-
-  // Pre-filter by first_prompt
-  const fp = (session.first_prompt || '').trim();
-  if (fp && SKIP_PROMPT_PATTERNS.some(p => p.test(fp))) {
-    return { sid, skip: 'trivial prompt' };
-  }
-
-  // The file_path recorded in work-list.json may have disappeared between
-  // scan and extract (user cleanup, Claude Code retention, project rename).
-  // Treat that as a per-session error so the batch keeps going instead of
-  // the format-session subprocess throwing out of prepareSession().
-  if (!fs.existsSync(session.file_path)) {
-    return { sid, error: 'session file not found (removed after scan?)' };
-  }
-
-  // Format
-  const outPath = path.join(os.tmpdir(), `session-${sid}.txt`);
-  let meta;
-  try {
-    meta = formatSession(session.file_path, outPath);
-  } catch (e) {
-    return { sid, error: `format threw: ${(e.message || String(e)).split('\n')[0]}` };
-  }
-  if (!meta) {
-    return { sid, error: 'format failed' };
-  }
-
-  const formattedFiles = meta.output_files || [outPath];
-
-  // Pre-filter by size
-  let totalChars = 0;
-  for (const f of formattedFiles) {
-    try { totalChars += fs.statSync(f).size; } catch {}
-  }
-  if (totalChars < MIN_FORMATTED_CHARS) {
-    for (const f of formattedFiles) { try { fs.unlinkSync(f); } catch {} }
-    return { sid, skip: `${totalChars} chars < ${MIN_FORMATTED_CHARS} min` };
-  }
-
-  return { sid, ready: true, formattedFiles };
-}
-
-// ---------------------------------------------------------------------------
-// Parse Sonnet output and collect result
-// ---------------------------------------------------------------------------
-
-function parseResult(sid, output, verbose) {
-  const match = output.match(/SKILLS_EXTRACTED:\s*(\d+)\s*\(E:(\d+)\s+S:(\d+)\s+P:(\d+)\)/);
-  if (match) {
-    const [, total, e, s, p] = match.map(Number);
-    return { sid, total, episodic: e, semantic: s, procedural: p };
-  }
-  if (isCached(sid)) {
-    return { sid, total: 0, episodic: 0, semantic: 0, procedural: 0, note: 'cached, no count' };
-  }
-  if (verbose) {
-    const lastLines = output.split('\n').slice(-5).join('\n');
-    console.error(`  [${sid}] Last output:`, lastLines);
-  }
-  return null;
-}
-
-// ---------------------------------------------------------------------------
-// Main (async, batched parallel)
-// ---------------------------------------------------------------------------
-
-async function main() {
-  const platform = parsePlatformFlag();
-  const runner = createRunner(platform);
-  const opts = parseArgs();
-
-  if (!opts.workListPath) {
-    console.error('Usage: extract-skills.js <work-list.json> [options]');
-    process.exit(1);
-  }
-
-  const workList = JSON.parse(fs.readFileSync(opts.workListPath, 'utf-8'));
-  let sessions = workList.sessions || [];
-
-  if (opts.sessionIds) {
-    const idSet = new Set(opts.sessionIds);
-    sessions = sessions.filter(s => idSet.has(s.session_id));
-  }
-
-  console.log(`\nSessions: ${sessions.length} total`);
-  console.log(`Concurrency: ${opts.concurrency} parallel Sonnet calls\n`);
-
-  // Phase 1: Pre-filter and format all sessions (fast, serial)
-  // Segment-level caching is checked later in Phase 2
-  console.log('Phase 1: Pre-filtering and formatting...');
-  const prepared = [];
-  const totals = { episodic: 0, semantic: 0, procedural: 0, errors: 0, skipped: 0 };
-
-  for (const session of sessions) {
-    const result = prepareSession(session);
-    if (result.skip) {
-      console.log(`  ${result.sid} → SKIP (${result.skip})`);
-      totals.skipped++;
-    } else if (result.error) {
-      console.log(`  ${result.sid} → ERROR (${result.error})`);
-      totals.errors++;
-    } else {
-      prepared.push(result);
-      if (prepared.length >= opts.batchSize) break; // batch limit on real sessions only
-    }
-  }
-
-  const batch = sessions;
-  // Sessions that passed pre-filter but didn't fit in batch-size
-  const remaining = sessions.length - totals.skipped - totals.errors - prepared.length;
-
-  console.log(`  ${prepared.length} sessions ready, ${totals.skipped} skipped, ${totals.errors} errors\n`);
-
-  if (prepared.length === 0) {
-    console.log('No sessions to extract. Done.');
-    process.exit(0);
-  }
-
-  // Phase 2: Build work units (one per segment, not per session)
-  // Multi-segment sessions produce multiple work units, all processed in parallel
-  // Segment-level caching: skip segments that already have cached skills
-  const workUnits = [];
-  let segmentsCached = 0;
-  for (const { sid, formattedFiles } of prepared) {
-    const totalSegs = formattedFiles.length;
-    for (let si = 0; si < totalSegs; si++) {
-      const segInfo = totalSegs > 1 ? { index: si + 1, total: totalSegs } : null;
-      if (isSegmentCached(sid, segInfo)) {
-        segmentsCached++;
-        // Clean up formatted file for cached segment
-        try { fs.unlinkSync(formattedFiles[si]); } catch {}
-        continue;
-      }
-      // Find the original session object to pass to buildPrompt
-      const session = batch.find(s => s.session_id === sid);
-      const prompt = buildPrompt(formattedFiles[si]);
-      const projectName = deriveProjectName(session, opts) || 'unknown';
-      const projectSlug = opts.projectSlug || slugify(projectName) || 'unknown';
-      const projectMeta = {
-        projectName, projectSlug,
-        domain: opts.domain || 'computer-science',
-        subdomain: opts.subdomain || 'general',
-        contributor: opts.contributor || 'anonymous',
-        sessionId: sid,
-        extractedAt: new Date().toISOString().split('T')[0],
-      };
-      workUnits.push({ sid, prompt, segFile: formattedFiles[si], segInfo, projectMeta });
-    }
-  }
-
-  if (segmentsCached > 0) {
-    console.log(`  ${segmentsCached} segments already cached, skipped`);
-  }
-
-  if (workUnits.length === 0) {
-    console.log('All segments already cached. Nothing to do.');
-    process.exit(0);
-  }
-
-  console.log(`\nPhase 2: Extracting skills (${prepared.length} sessions → ${workUnits.length} Sonnet calls, ${opts.concurrency} at a time)...`);
-  const results = [];
-  let completed = 0;
-
-  for (let i = 0; i < workUnits.length; i += opts.concurrency) {
-    const chunk = workUnits.slice(i, i + opts.concurrency);
-    const batchNum = Math.floor(i / opts.concurrency) + 1;
-    const totalBatches = Math.ceil(workUnits.length / opts.concurrency);
-
-    console.log(`\n  Batch ${batchNum}/${totalBatches} (${chunk.length} calls):`);
-
-    const promises = chunk.map(async ({ sid, prompt, segFile, segInfo, projectMeta }) => {
-      const segLabel = segInfo ? ` seg${segInfo.index}/${segInfo.total}` : '';
-      const startTime = Date.now();
-      const { ok, output, error } = await runner.extract(prompt, 300_000);
-      const elapsed = ((Date.now() - startTime) / 1000).toFixed(1);
-
-      // Clean up segment file
-      try { fs.unlinkSync(segFile); } catch {}
-
-      if (!ok) {
-        console.log(`    ${sid}${segLabel} → ERROR (${error}) [${elapsed}s]`);
-        totals.errors++;
-        return null;
-      }
-
-      // Parse skill-md blocks from output → inject project metadata → write files → validate
-      const written = writeAndValidateSkills(sid, output, projectMeta, segInfo);
-
-      const result = parseResult(sid, output, opts.verbose);
-      if (result) {
-        totals.episodic += result.episodic;
-        totals.semantic += result.semantic;
-        totals.procedural += result.procedural;
-        console.log(`    ${sid}${segLabel} → ${result.total} skills (E:${result.episodic} S:${result.semantic} P:${result.procedural}) [${elapsed}s]`);
-        return result;
-      } else if (written > 0) {
-        console.log(`    ${sid}${segLabel} → ${written} skills written [${elapsed}s]`);
-        return { sid, total: written, episodic: 0, semantic: 0, procedural: 0 };
-      } else {
-        console.log(`    ${sid}${segLabel} → 0 skills [${elapsed}s]`);
-        return null;
-      }
-    });
-
-    const chunkResults = await Promise.all(promises);
-    results.push(...chunkResults.filter(Boolean));
-    completed += chunk.length;
-    console.log(`  Progress: ${completed}/${workUnits.length}`);
-
-    if (opts.singleBatch) {
-      const remainingCalls = workUnits.length - completed;
-      console.log(`\n  --single-batch: stopping after batch ${batchNum}. ${remainingCalls} Sonnet calls remaining.`);
-      break;
-    }
-  }
-
-  // Clean up any remaining formatted files
-  for (const { formattedFiles } of prepared) {
-    for (const f of formattedFiles) { try { fs.unlinkSync(f); } catch {} }
-  }
-
-  // Count actual cached skills by reading cache directories
-  const cachedCounts = { total: 0, episodic: 0, semantic: 0, procedural: 0 };
-  const processedSids = new Set(prepared.map(p => p.sid));
-  for (const sid of processedSids) {
-    const cacheDir = path.join(os.homedir(), '.researchskills', 'cache', 'skills', sid);
-    try {
-      const files = fs.readdirSync(cacheDir).filter(f => f.endsWith('.md'));
-      for (const file of files) {
-        const content = fs.readFileSync(path.join(cacheDir, file), 'utf-8');
-        const typeMatch = content.match(/memory_type:\s*(\w+)/);
-        const memType = typeMatch ? typeMatch[1] : null;
-        cachedCounts.total++;
-        if (memType === 'episodic') cachedCounts.episodic++;
-        else if (memType === 'semantic') cachedCounts.semantic++;
-        else if (memType === 'procedural') cachedCounts.procedural++;
-      }
-    } catch {}
-  }
-
-  // Sonnet-reported totals (what Sonnet claimed to extract)
-  const sonnetTotal = totals.episodic + totals.semantic + totals.procedural;
-
-  console.log(`
-═══════════════════════════════════════════════
-  Extraction Complete
-═══════════════════════════════════════════════
-  Sessions processed: ${prepared.length}
-  Sonnet reported:     ${sonnetTotal} skills (E:${totals.episodic} S:${totals.semantic} P:${totals.procedural})
-  Actually cached:    ${cachedCounts.total} skills (E:${cachedCounts.episodic} S:${cachedCounts.semantic} P:${cachedCounts.procedural})
-  Skipped:            ${totals.skipped} (trivial/too short)
-  Errors:             ${totals.errors}
-${remaining > 0 ? `  Remaining:          ${remaining} (run again to continue)` : ''}
-═══════════════════════════════════════════════`);
-
-  const summaryPath = path.join(os.tmpdir(), 'extract-skills-summary.json');
-  fs.writeFileSync(summaryPath, JSON.stringify({
-    batch_size: prepared.length,
-    model_reported: sonnetTotal,
-    cached: cachedCounts,
-    totals,
-    remaining,
-    results,
-  }, null, 2));
-  console.log(`\nSummary: ${summaryPath}`);
-}
-
-main().catch(err => {
-  console.error('Fatal:', err);
-  process.exit(1);
-});
diff --git a/researchskills-extract/scripts/finalize.js b/researchskills-extract/scripts/finalize.js
deleted file mode 100644
index b7bfe9a..0000000
--- a/researchskills-extract/scripts/finalize.js
+++ /dev/null
@@ -1,175 +0,0 @@
-#!/usr/bin/env node
-/**
- * finalize.js
- *
- * Project finalizer for /researchskills-extract. Given a project metadata file
- * describing which cached skills belong to one project, this script:
- *
- *   1. Collects validated skills  (validate-skills.js collect)
- *   2. Uploads to researchskills.ai  (upload-skills.js)
- *   3. Cleans up temp dir
- *
- * Usage:
- *   finalize.js <project-meta.json> [--no-upload] [--no-open]
- *
- * project-meta.json shape:
- *   {
- *     "project_slug":   "my-project",
- *     "session_ids":    ["id1", "id2", ...],
- *     "anchor":         { "type": "project", "project_name": "...", "project_description": "..." },
- *     "domain":         "physics",
- *     "subdomain":      "computational-physics",
- *     "contributor":    "Name (Institution)",
- *     "is_test":        false
- *   }
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const { execFileSync } = require('child_process');
-
-const UTILS_DIR = path.join(os.homedir(), '.claude', 'utils');
-const VALIDATE_SKILLS = path.join(UTILS_DIR, 'validate-skills.js');
-const UPLOAD_SKILLS = path.join(UTILS_DIR, 'upload-skills.js');
-
-function runNode(script, args, opts = {}) {
-  return execFileSync('node', [script, ...args], {
-    encoding: 'utf-8',
-    stdio: opts.inherit ? 'inherit' : ['ignore', 'pipe', 'inherit'],
-  });
-}
-
-function slugify(s) {
-  return String(s || 'project')
-    .toLowerCase()
-    .replace(/[^a-z0-9]+/g, '-')
-    .replace(/^-+|-+$/g, '')
-    .substring(0, 60) || 'project';
-}
-
-function finalize(metaPath, options = {}) {
-  const meta = JSON.parse(fs.readFileSync(metaPath, 'utf-8'));
-  const slug = slugify(meta.project_slug || (meta.anchor && meta.anchor.project_name));
-  const tmp = os.tmpdir();
-  const outputDir = path.join(tmp, `${slug}-skills`);
-
-  const sessionIds = meta.session_ids || [];
-  if (sessionIds.length === 0) {
-    throw new Error('project-meta.json has empty session_ids');
-  }
-
-  // 1. Collect validated skills
-  runNode(VALIDATE_SKILLS, ['collect', outputDir, sessionIds.join(',')]);
-  console.log(`✓ Skills collected → ${outputDir}`);
-
-  if (!options.upload) {
-    console.log(`⚠ Skills collected but NOT uploaded (no --upload flag).`);
-    console.log(`  Review the skills locally at: ${outputDir}`);
-    console.log(`  To upload, re-run with --upload`);
-    return { outputDir, uploaded: false };
-  }
-
-  // 2. Upload skills
-  const uploadArgs = [outputDir];
-  if (options.headless) uploadArgs.push('--headless');
-  if (options.consent) uploadArgs.push('--consent');
-  if (options.noOpen) uploadArgs.push('--no-open');
-  if (meta.project_slug) {
-    uploadArgs.push('--project-slug', meta.project_slug);
-  }
-  if (meta.anchor && meta.anchor.project_name) {
-    uploadArgs.push('--project-name', meta.anchor.project_name);
-  }
-  try {
-    const out = runNode(UPLOAD_SKILLS, uploadArgs);
-    process.stdout.write(out);
-    return { outputDir, uploaded: true };
-  } catch (err) {
-    console.error('⚠ Upload step failed (see output above)');
-    return { outputDir, uploaded: false };
-  } finally {
-    // 3. Clean up temp dir
-    try {
-      fs.rmSync(outputDir, { recursive: true, force: true });
-    } catch (_) {
-      // best-effort cleanup
-    }
-  }
-}
-
-// CLI
-if (require.main === module) {
-  const args = process.argv.slice(2);
-
-  // Parse CLI args — supports both JSON file and inline flags
-  let metaPath = null;
-  const cliOpts = {};
-  for (let i = 0; i < args.length; i++) {
-    switch (args[i]) {
-      case '--domain':         cliOpts.domain = args[++i]; break;
-      case '--subdomain':      cliOpts.subdomain = args[++i]; break;
-      case '--contributor':    cliOpts.contributor = args[++i]; break;
-      case '--session-ids':    cliOpts.sessionIds = args[++i].split(','); break;
-      case '--project-name':   cliOpts.projectName = args[++i]; break;
-      case '--project-slug':   cliOpts.projectSlug = args[++i]; break;
-      case '--description':    cliOpts.description = args[++i]; break;
-      case '--test':           cliOpts.isTest = true; break;
-      case '--upload':          cliOpts.upload = true; break;
-      case '--no-upload':      cliOpts.noUpload = true; break;
-      case '--no-open':        cliOpts.noOpen = true; break;
-      case '--headless':       cliOpts.headless = true; break;
-      case '--consent':        cliOpts.consent = true; break;
-      default:
-        if (!args[i].startsWith('-') && !metaPath) metaPath = args[i];
-    }
-  }
-
-  // Build meta from CLI args or read from JSON file
-  let resolvedMetaPath;
-  if (metaPath) {
-    resolvedMetaPath = path.resolve(metaPath);
-    if (!fs.existsSync(resolvedMetaPath)) {
-      console.error(`Error: not found: ${resolvedMetaPath}`);
-      process.exit(1);
-    }
-  } else if (cliOpts.sessionIds && cliOpts.sessionIds.length > 0) {
-    // Build meta JSON from CLI flags and write to temp file
-    const name = cliOpts.projectName || 'project';
-    const meta = {
-      project_slug: cliOpts.projectSlug || slugify(name),
-      session_ids: cliOpts.sessionIds,
-      anchor: {
-        type: 'project',
-        project_name: name,
-        project_description: cliOpts.description || '',
-      },
-      domain: cliOpts.domain || 'computer-science',
-      subdomain: cliOpts.subdomain || 'general',
-      contributor: cliOpts.contributor || 'anonymous',
-      is_test: cliOpts.isTest || false,
-    };
-    resolvedMetaPath = path.join(os.tmpdir(), `finalize-meta-${Date.now()}.json`);
-    fs.writeFileSync(resolvedMetaPath, JSON.stringify(meta, null, 2));
-  } else {
-    console.error('Usage: finalize.js <project-meta.json> [--no-upload] [--no-open]');
-    console.error('   or: finalize.js --session-ids id1,id2 --domain ... --project-name ...');
-    process.exit(1);
-  }
-
-  try {
-    finalize(resolvedMetaPath, {
-      upload: cliOpts.upload && !cliOpts.noUpload,
-      noOpen: cliOpts.noOpen || false,
-      headless: cliOpts.headless || false,
-      consent: cliOpts.consent || false,
-    });
-  } catch (err) {
-    console.error(`Error: ${err.message}`);
-    process.exit(1);
-  }
-}
-
-module.exports = { finalize, slugify };
diff --git a/researchskills-extract/scripts/format-session.js b/researchskills-extract/scripts/format-session.js
deleted file mode 100644
index a021e5e..0000000
--- a/researchskills-extract/scripts/format-session.js
+++ /dev/null
@@ -1,832 +0,0 @@
-#!/usr/bin/env node
-/**
- * format-session.js
- *
- * Preprocess a raw Claude Code or Codex .jsonl session file, or a
- * Gemini brain entry directory, into compact text optimized for
- * research skill extraction.
- *
- * Supports Claude Code JSONL, Codex JSONL, and Gemini brain formats.
- *
- * Design principle: We're extracting HUMAN tacit knowledge. The human's
- * inputs, corrections, and decisions are the signal. AI outputs and tool
- * mechanics are context at best, noise at worst.
- *
- * Optimizations:
- *   - Strip IDE context tags, slash commands, local commands from USER msgs
- *   - Mark context continuations as [CONTEXT] not USER
- *   - Drop git-related USER messages (commit/push instructions)
- *   - Extract tool_use parameters (file paths) → [Read field_equations.py]
- *   - ASSISTANT truncated to 80 chars (just enough for continuity)
- *   - Bash errors get more space (120), normal output less (60)
- *   - Git tool output dropped entirely
- *   - "rejected" preserved as high-value signal
- *   - Consecutive low-value tool lines collapsed/dropped
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-
-const USER_MAX_CHARS = 1000;
-const ASSISTANT_MAX_CHARS = 80;
-const BASH_ERROR_MAX_CHARS = 120;
-const BASH_NORMAL_MAX_CHARS = 60;
-const SEGMENT_THRESHOLD = 200000;
-const SEGMENT_SIZE = 100000;
-
-function truncate(text, max) {
-  if (!text) return '';
-  const str = String(text);
-  if (str.length <= max) return str;
-  return str.substring(0, max) + '…';
-}
-
-function basename(filepath) {
-  if (!filepath) return null;
-  const parts = String(filepath).split('/');
-  return parts[parts.length - 1] || null;
-}
-
-// ---------------------------------------------------------------------------
-// Detect session format
-// ---------------------------------------------------------------------------
-
-function detectFormat(lines) {
-  for (const line of lines.slice(0, 5)) {
-    const trimmed = line.trim();
-    if (!trimmed) continue;
-    try {
-      const entry = JSON.parse(trimmed);
-      if (entry.type === 'session_meta') return 'codex';
-      if (entry.type === 'event_msg') return 'codex';
-      if (entry.type === 'response_item') return 'codex';
-      if (entry.type === 'turn_context') return 'codex';
-      if (entry.type === 'user' || entry.type === 'assistant') return 'claude';
-      if (entry.message && entry.message.role) return 'claude';
-    } catch { continue; }
-  }
-  return 'claude'; // default fallback
-}
-
-// ---------------------------------------------------------------------------
-// Pass 0: Parse JSONL into structured messages
-// ---------------------------------------------------------------------------
-
-// Extract tool_use params (file_path, command) for richer context — Claude Code
-function extractToolInfo(item) {
-  if (item.type !== 'tool_use' || !item.name) return null;
-  const info = { name: item.name };
-  const inp = item.input || {};
-  if (inp.file_path) info.file = basename(inp.file_path);
-  if (inp.command) info.cmd = truncate(inp.command, 60);
-  if (inp.pattern) info.pattern = truncate(inp.pattern, 40);
-  if (inp.prompt && item.name === 'Agent') info.desc = truncate(inp.description || inp.prompt, 40);
-  return info;
-}
-
-function extractContentParts(content) {
-  if (typeof content === 'string') return { parts: [content], toolInfos: [] };
-  if (!Array.isArray(content)) return { parts: [], toolInfos: [] };
-  const parts = [];
-  const toolInfos = [];
-  for (const item of content) {
-    if (!item) continue;
-    if (typeof item === 'string') {
-      parts.push(item);
-    } else if (
-      (item.type === 'text' || item.type === 'input_text' || item.type === 'output_text') &&
-      item.text
-    ) {
-      parts.push(item.text);
-    } else if (item.type === 'tool_use' && item.name) {
-      const info = extractToolInfo(item);
-      if (info) toolInfos.push(info);
-      let label = `[Tool: ${item.name}`;
-      if (info && info.file) label += ` ${info.file}`;
-      else if (info && info.cmd) label += ` ${info.cmd}`;
-      else if (info && info.pattern) label += ` ${info.pattern}`;
-      label += ']';
-      parts.push(label);
-    } else if (item.type === 'tool_result') {
-      let resText = '';
-      if (typeof item.content === 'string') {
-        resText = item.content;
-      } else if (Array.isArray(item.content)) {
-        resText = item.content
-          .map((c) => (typeof c === 'string' ? c : c.text || ''))
-          .join(' ');
-      }
-      if (resText) parts.push(`[Tool Result: ${resText}]`);
-    }
-  }
-  return { parts, toolInfos };
-}
-
-// Claude Code message extractor
-function extractMessageClaude(entry) {
-  if (!entry) return null;
-  if (entry.payload && entry.payload.type && entry.payload.type !== 'message') return null;
-  const msg = entry.message || entry.payload || entry;
-  const role = msg.role || entry.role ||
-    (entry.type === 'user' || entry.type === 'assistant' ? entry.type : null);
-  if (role !== 'user' && role !== 'assistant') return null;
-  const { parts, toolInfos } = extractContentParts(msg.content);
-  if (parts.length === 0) return null;
-  const text = parts.join(' ').replace(/\s+/g, ' ').trim();
-  if (!text) return null;
-  const timestamp = entry.timestamp || entry.ts || msg.timestamp || null;
-  return { role, text, timestamp, toolInfos };
-}
-
-// ---------------------------------------------------------------------------
-// Codex message extractor
-// ---------------------------------------------------------------------------
-
-// Parse Codex exec_command output: strip metadata prefix, extract exit code
-function parseExecOutput(output) {
-  if (!output || typeof output !== 'string') return { exitCode: 0, content: String(output || '') };
-  const lines = output.split('\n');
-  let exitCode = 0;
-  let outputStartIdx = 0;
-
-  for (let i = 0; i < Math.min(lines.length, 10); i++) {
-    const line = lines[i];
-    const exitMatch = line.match(/^Process exited with code (\d+)/);
-    if (exitMatch) {
-      exitCode = parseInt(exitMatch[1], 10);
-    }
-    if (line.startsWith('Output:')) {
-      outputStartIdx = i + 1;
-      break;
-    }
-  }
-
-  const content = lines.slice(outputStartIdx).join('\n').trim();
-  return { exitCode, content };
-}
-
-// Classify Codex exec_command by cmd string
-function classifyExecCommand(cmd) {
-  if (!cmd) return 'shell';
-  const trimmed = cmd.trim();
-  if (/^(cat|head|tail|less|bat)\s/.test(trimmed)) return 'read';
-  if (/^(grep|rg|ag)\s/.test(trimmed)) return 'search';
-  if (/^(find|fd|ls)\s/.test(trimmed)) return 'discover';
-  if (/^git\s/.test(trimmed)) return 'git';
-  return 'shell';
-}
-
-// Extract file target from exec_command cmd string
-function extractCmdTarget(cmd) {
-  if (!cmd) return null;
-  // cat/head/tail <file>
-  const readMatch = cmd.match(/^(?:cat|head|tail|less|bat)\s+(?:-[^\s]+\s+)*(\S+)/);
-  if (readMatch) return basename(readMatch[1]);
-  return null;
-}
-
-function extractMessageCodex(entry) {
-  if (!entry) return null;
-  const ts = entry.timestamp || null;
-
-  // event_msg — user message
-  if (entry.type === 'event_msg' && entry.payload) {
-    if (entry.payload.type === 'user_message') {
-      const text = (entry.payload.message || '').trim();
-      if (!text) return null;
-      return { role: 'user', text, timestamp: ts, toolInfos: [] };
-    }
-    if (entry.payload.type === 'agent_message') {
-      const text = (entry.payload.message || '').trim();
-      if (!text) return null;
-      return { role: 'assistant', text, timestamp: ts, toolInfos: [] };
-    }
-    // Skip token_count, task_started, task_complete
-    return null;
-  }
-
-  // response_item — assistant message
-  if (entry.type === 'response_item' && entry.payload) {
-    const p = entry.payload;
-
-    // Assistant text message
-    if (p.type === 'message' && p.role === 'assistant') {
-      const parts = [];
-      if (Array.isArray(p.content)) {
-        for (const c of p.content) {
-          if (c && c.type === 'output_text' && c.text) parts.push(c.text);
-        }
-      }
-      const text = parts.join(' ').replace(/\s+/g, ' ').trim();
-      if (!text) return null;
-      return { role: 'assistant', text, timestamp: ts, toolInfos: [] };
-    }
-
-    // Function call (tool use)
-    if (p.type === 'function_call') {
-      const name = p.name || 'unknown';
-      let args = {};
-      try { args = JSON.parse(p.arguments || '{}'); } catch {}
-
-      // Build tool label similar to Claude Code format
-      let label;
-      if (name === 'exec_command') {
-        const cmd = args.cmd || '';
-        const cmdType = classifyExecCommand(cmd);
-        const target = extractCmdTarget(cmd);
-        if (cmdType === 'read' && target) {
-          label = `[Tool: Read ${target}]`;
-        } else if (cmdType === 'git') {
-          label = `[Tool: Bash git ${truncate(cmd, 40)}]`;
-        } else {
-          label = `[Tool: Bash ${truncate(cmd, 60)}]`;
-        }
-      } else if (name === 'apply_patch') {
-        // Extract filename from patch
-        const patchInput = p.input || args.patch || '';
-        const fileMatch = patchInput.match(/\*\*\* (?:Add|Update|Delete) File:\s*(\S+)/);
-        const file = fileMatch ? basename(fileMatch[1]) : null;
-        label = file ? `[Tool: Edit ${file}]` : '[Tool: Edit]';
-      } else {
-        label = `[Tool: ${name}]`;
-      }
-
-      return {
-        role: 'assistant',
-        text: label,
-        timestamp: ts,
-        toolInfos: [{ name: name === 'exec_command' ? 'Bash' : name }],
-      };
-    }
-
-    // Function call output (tool result)
-    if (p.type === 'function_call_output') {
-      const rawOutput = typeof p.output === 'string' ? p.output : JSON.stringify(p.output || '');
-      const parsed = parseExecOutput(rawOutput);
-      const content = parsed.content || rawOutput;
-      return {
-        role: 'user',
-        text: `[Tool Result: ${content}]`,
-        timestamp: ts,
-        toolInfos: [],
-        _exitCode: parsed.exitCode,
-      };
-    }
-
-    // custom_tool_call (apply_patch)
-    if (p.type === 'custom_tool_call') {
-      if (p.name === 'apply_patch') {
-        const status = p.status === 'completed' ? 'ok' : (p.status || 'unknown');
-        const fileMatch = (p.input || '').match(/\*\*\* (?:Add|Update|Delete) File:\s*(\S+)/);
-        const file = fileMatch ? basename(fileMatch[1]) : null;
-        const label = file ? `[Tool: Edit ${file} → ${status}]` : `[Tool: Edit → ${status}]`;
-        return { role: 'tool', text: label, timestamp: ts, toolInfos: [] };
-      }
-      return null;
-    }
-
-    // Skip reasoning (encrypted)
-    if (p.type === 'reasoning') return null;
-  }
-
-  // Skip session_meta, turn_context
-  return null;
-}
-
-// Unified message extractor
-function extractMessage(entry, format) {
-  if (format === 'codex') return extractMessageCodex(entry);
-  return extractMessageClaude(entry);
-}
-
-// ---------------------------------------------------------------------------
-// Pass 1.5: Clean USER messages (strip IDE tags, commands, git noise)
-// ---------------------------------------------------------------------------
-
-const GIT_USER_PATTERNS = [
-  /^(嗯\s*)?commit\s*(and\s*push)?/i,
-  /^push/i,
-  /^git\s/i,
-];
-
-function cleanUserMessage(text) {
-  let t = text;
-  t = t.replace(/<ide_opened_file>.*?<\/ide_opened_file>\s*/g, '').trim();
-  t = t.replace(/<ide_selection>.*?<\/ide_selection>\s*/g, '').trim();
-  if (/^<command-name>/.test(t)) return null;
-  if (/^<local-command/.test(t)) return null;
-  if (!t) return null;
-  return t;
-}
-
-function classifyUserMessage(text) {
-  if (/^This session is being continued from a previous conversation/.test(text)) {
-    return 'context';
-  }
-  if (GIT_USER_PATTERNS.some(p => p.test(text.trim()))) {
-    return 'git';
-  }
-  return 'user';
-}
-
-// ---------------------------------------------------------------------------
-// Pass 2: Merge tool calls with results, compress
-// ---------------------------------------------------------------------------
-
-const DROP_TOOLS = new Set([
-  'ToolSearch', 'TodoWrite', 'Glob', 'ExitPlanMode', 'EnterPlanMode',
-  'Skill', 'Monitor', 'ScheduleWakeup', 'CronCreate', 'CronDelete',
-]);
-
-const GIT_OUTPUT_PATTERNS = [
-  /^\[main [0-9a-f]/,
-  /^On branch /,
-  /^To https?:\/\//,
-  /^remote: /,
-  /^branch '.+' set up to track/,
-  /^Enumerating objects/,
-  /^Your branch is /,
-  /^Changes (not staged|to be committed)/,
-  /^(Untracked files|no changes added)/,
-  /^\s*modified:/,
-  /^\s*new file:/,
-  /^\s*deleted:/,
-  /^diff --git/,
-  /^[0-9a-f]{7,} /,
-];
-
-function isGitOutput(text) {
-  const first = text.trim().split('\n')[0] || '';
-  return GIT_OUTPUT_PATTERNS.some(p => p.test(first));
-}
-
-function compressToolResult(text, toolName, toolInfo) {
-  if (DROP_TOOLS.has(toolName)) return null;
-  const clean = text.trim();
-
-  if (toolName === 'Read') {
-    if (/error|not found|does not exist/i.test(clean.substring(0, 80))) {
-      return truncate(clean, BASH_ERROR_MAX_CHARS);
-    }
-    return null;
-  }
-
-  if (toolName === 'Edit' || toolName === 'Write') {
-    if (/doesn't want to proceed/.test(clean)) return 'rejected';
-    if (/error|<tool_use_error>/i.test(clean.substring(0, 80))) return 'error';
-    return null;
-  }
-
-  if (toolName === 'Grep') return null;
-
-  if (toolName === 'Agent') {
-    if (/doesn't want to proceed/.test(clean)) return 'rejected';
-    return truncate(clean, BASH_NORMAL_MAX_CHARS);
-  }
-
-  if (toolName === 'AskUserQuestion') {
-    return truncate(clean, 200);
-  }
-
-  if (toolName === 'Bash') {
-    if (/^\(Bash completed with no output\)$/.test(clean)) return null;
-    if (isGitOutput(clean)) return null;
-    if (/Exit code [^0]|Error|error|FAIL|panic|Traceback|ENOENT|EACCES/.test(clean.substring(0, 200))) {
-      return truncate(clean, BASH_ERROR_MAX_CHARS);
-    }
-    return truncate(clean, BASH_NORMAL_MAX_CHARS);
-  }
-
-  if (/^\(Bash completed/.test(clean)) return null;
-  if (/^Wasted call/.test(clean)) return null;
-  if (/^\[Request interrupted/.test(clean)) return null;
-  if (/^Todos have been modified/.test(clean)) return null;
-  if (/doesn't want to proceed/.test(clean)) return 'rejected';
-  if (/approved your plan/.test(clean)) return 'plan approved';
-  if (/answered your questions/.test(clean)) return truncate(clean, 200);
-  return truncate(clean, BASH_NORMAL_MAX_CHARS);
-}
-
-function isToolCall(msg) {
-  return msg.role === 'assistant' && /^\[Tool: \w+/.test(msg.text);
-}
-
-function getToolName(msg) {
-  const m = msg.text.match(/^\[Tool: (\w+)/);
-  return m ? m[1] : null;
-}
-
-function getToolFile(msg) {
-  const m = msg.text.match(/^\[Tool: \w+ (.+)\]$/);
-  return m ? m[1] : null;
-}
-
-function isToolResult(msg) {
-  return msg.role === 'user' && msg.text.startsWith('[Tool Result:');
-}
-
-function getToolResultContent(msg) {
-  const m = msg.text.match(/^\[Tool Result: ([\s\S]*)\]$/);
-  return m ? m[1] : msg.text;
-}
-
-function mergeAndCompress(messages) {
-  const output = [];
-  let i = 0;
-
-  while (i < messages.length) {
-    const msg = messages[i];
-
-    // Codex custom_tool_call results are already role:'tool' — pass through
-    if (msg.role === 'tool') {
-      output.push(msg);
-      i++;
-      continue;
-    }
-
-    if (isToolCall(msg)) {
-      const toolName = getToolName(msg);
-      const toolFile = getToolFile(msg);
-      const ts = msg.timestamp;
-
-      let resultMsg = null;
-      let lookAhead = i + 1;
-      while (lookAhead < messages.length && lookAhead <= i + 2) {
-        if (isToolResult(messages[lookAhead])) {
-          resultMsg = messages[lookAhead];
-          break;
-        }
-        if (messages[lookAhead].text.includes('[Request interrupted')) {
-          lookAhead++;
-          continue;
-        }
-        break;
-      }
-
-      if (resultMsg) {
-        if (toolName === 'Bash' && toolFile && /^git\s/.test(toolFile)) {
-          i = lookAhead + 1;
-          continue;
-        }
-
-        const resultContent = getToolResultContent(resultMsg);
-        const compressed = compressToolResult(resultContent, toolName);
-
-        if (compressed === null && !toolFile) {
-          i = lookAhead + 1;
-          continue;
-        }
-
-        const isRejected = compressed === 'rejected';
-
-        let label;
-        if (compressed && compressed !== 'rejected' && compressed !== 'error') {
-          label = toolFile
-            ? `[${toolName} ${toolFile} → ${compressed}]`
-            : `[${toolName} → ${compressed}]`;
-        } else if (compressed === 'rejected' || compressed === 'error') {
-          label = toolFile
-            ? `[${toolName} ${toolFile} → ${compressed}]`
-            : `[${toolName} → ${compressed}]`;
-        } else {
-          label = `[${toolName} ${toolFile}]`;
-        }
-
-        output.push({ role: 'tool', text: label, timestamp: ts, highValue: isRejected });
-        i = lookAhead + 1;
-        continue;
-      }
-
-      if (DROP_TOOLS.has(toolName)) { i++; continue; }
-      if (toolName === 'Bash' && toolFile && /^git\s/.test(toolFile)) { i++; continue; }
-      const label = toolFile ? `[${toolName} ${toolFile}]` : `[${toolName}]`;
-      output.push({ role: 'tool', text: label, timestamp: ts });
-      i++;
-      continue;
-    }
-
-    if (isToolResult(msg)) { i++; continue; }
-    if (msg.text.includes('[Request interrupted')) { i++; continue; }
-
-    output.push(msg);
-    i++;
-  }
-
-  return output;
-}
-
-// ---------------------------------------------------------------------------
-// Pass 3: Collapse consecutive low-value tool lines
-// ---------------------------------------------------------------------------
-
-function isLowValueTool(msg) {
-  if (msg.highValue) return false;
-  if (/^\[\w+( [\w._-]+)?\]$/.test(msg.text)) return true;
-  if (/^\[\w+ → .{1,20}\]$/.test(msg.text)) return true;
-  if (/^\[\d+x \w+\]$/.test(msg.text)) return true;
-  return false;
-}
-
-function collapseConsecutiveTools(messages) {
-  const output = [];
-  let i = 0;
-
-  while (i < messages.length) {
-    const msg = messages[i];
-
-    if (msg.role === 'tool' && isLowValueTool(msg)) {
-      let count = 0;
-      let lastTs = msg.timestamp;
-      while (i < messages.length && messages[i].role === 'tool' && isLowValueTool(messages[i])) {
-        const nxMatch = messages[i].text.match(/^\[(\d+)x /);
-        count += nxMatch ? parseInt(nxMatch[1], 10) : 1;
-        lastTs = messages[i].timestamp;
-        i++;
-      }
-      if (count >= 5) {
-        output.push({ role: 'tool', text: `[${count}x tools]`, timestamp: lastTs });
-      }
-      continue;
-    }
-
-    output.push(msg);
-    i++;
-  }
-
-  return output;
-}
-
-// ---------------------------------------------------------------------------
-// Pass 4: Filter assistant execution noise
-// ---------------------------------------------------------------------------
-
-const ASSISTANT_NOISE = [
-  /^Now /i, /^Let me /i, /^I'll /i, /^Done\b/i,
-  /^已/, /^好的/, /^Updated?\b/i, /^Created?\b/i,
-  /^Fixed?\b/i, /^Committed?\b/i, /^Pushed?\b/i,
-  /^推了/, /^推送了/, /^完成/,
-  /^\[Tool:/, /^No commits yet/,
-];
-
-function isAssistantNoise(text) {
-  const trimmed = text.trim();
-  if (trimmed.length < 25) return true;
-  return ASSISTANT_NOISE.some(p => p.test(trimmed));
-}
-
-function filterAssistantNoise(messages) {
-  return messages.filter(msg => {
-    if (msg.role !== 'assistant') return true;
-    return !isAssistantNoise(msg.text);
-  });
-}
-
-// ---------------------------------------------------------------------------
-// Main format pipeline
-// ---------------------------------------------------------------------------
-
-function formatSession(jsonlPath) {
-  const content = fs.readFileSync(jsonlPath, 'utf-8');
-  const lines = content.split('\n');
-
-  // Detect format from first few lines
-  const format = detectFormat(lines);
-
-  // Pass 1: parse all messages
-  const rawMessages = [];
-  let startTimestamp = null;
-  let messageCount = 0;
-
-  for (const line of lines) {
-    const trimmed = line.trim();
-    if (!trimmed) continue;
-    let entry;
-    try { entry = JSON.parse(trimmed); } catch { continue; }
-    const msg = extractMessage(entry, format);
-    if (!msg) continue;
-    if (!startTimestamp && msg.timestamp) startTimestamp = msg.timestamp;
-    rawMessages.push(msg);
-    messageCount += 1;
-  }
-
-  // Pass 1.5: clean USER messages
-  const cleaned = [];
-  for (const msg of rawMessages) {
-    if (msg.role === 'user' && !msg.text.startsWith('[Tool Result:')) {
-      const cls = classifyUserMessage(msg.text);
-      if (cls === 'git') continue;
-      if (cls === 'context') {
-        cleaned.push({ ...msg, role: 'context', text: truncate(msg.text, USER_MAX_CHARS) });
-        continue;
-      }
-      const cleanText = cleanUserMessage(msg.text);
-      if (!cleanText) continue;
-      cleaned.push({ ...msg, text: cleanText });
-    } else {
-      cleaned.push(msg);
-    }
-  }
-
-  // Pass 2: merge tool calls with results
-  const merged = mergeAndCompress(cleaned);
-
-  // Pass 3: collapse consecutive low-value tools
-  const collapsed = collapseConsecutiveTools(merged);
-
-  // Pass 4: filter assistant noise
-  const filtered = filterAssistantNoise(collapsed);
-
-  // Format final output
-  const formatted = filtered.map((msg) => {
-    const ts = msg.timestamp ? `[${msg.timestamp}] ` : '';
-    if (msg.role === 'tool') return `${ts}${msg.text}`;
-    if (msg.role === 'context') return `${ts}[CONTEXT]: ${truncate(msg.text, USER_MAX_CHARS)}`;
-    const maxLen = msg.role === 'user' ? USER_MAX_CHARS : ASSISTANT_MAX_CHARS;
-    return `${ts}${msg.role.toUpperCase()}: ${truncate(msg.text, maxLen)}`;
-  });
-
-  return {
-    text: formatted.join('\n'),
-    startTimestamp,
-    messageCount,
-  };
-}
-
-// ---------------------------------------------------------------------------
-// Gemini brain entry formatter
-// ---------------------------------------------------------------------------
-
-/**
- * Format a Gemini brain entry directory into extraction-ready text.
- *
- * Brain entries contain task.md, implementation_plan.md, walkthrough.md,
- * plus .resolved.N snapshots showing progression. We reconstruct a
- * session narrative from these artifacts.
- */
-function formatGeminiSession(dirPath) {
-  const lines = [];
-  let startTimestamp = null;
-  let messageCount = 0;
-
-  // Collect metadata timestamps and summaries, sorted by updatedAt for
-  // deterministic output across machines (readdirSync order varies).
-  const metaFiles = fs.readdirSync(dirPath).filter((f) => f.endsWith('.metadata.json'));
-  const metaEntries = [];
-  for (const mf of metaFiles) {
-    try {
-      const meta = JSON.parse(fs.readFileSync(path.join(dirPath, mf), 'utf-8'));
-      metaEntries.push({ updatedAt: meta.updatedAt || '', summary: meta.summary || '' });
-    } catch {}
-  }
-  metaEntries.sort((a, b) => a.updatedAt.localeCompare(b.updatedAt));
-  const timestamps = metaEntries.filter((e) => e.updatedAt).map((e) => e.updatedAt);
-  if (timestamps.length > 0) startTimestamp = timestamps[0];
-  for (const entry of metaEntries) {
-    if (entry.summary) {
-      lines.push(`[CONTEXT]: ${truncate(entry.summary, USER_MAX_CHARS)}`);
-      messageCount += 1;
-    }
-  }
-
-  // Read resolved snapshots first, sorted by file mtime for correct chronology.
-  // Gemini uses independent .resolved.N counters per artifact, so sorting
-  // by mtime across all artifacts preserves the actual decision sequence.
-  // Current task.md/plan are emitted AFTER snapshots since snapshots are
-  // older states and current files represent the latest state.
-  const allDirFiles = fs.readdirSync(dirPath);
-  const resolvedSnapshots = [];
-  for (const f of allDirFiles) {
-    const m = f.match(/^(.+)\.resolved\.(\d+)$/);
-    if (!m) continue;
-    const artifact = m[1];
-    const fullPath = path.join(dirPath, f);
-    let mtime = 0;
-    try { mtime = fs.statSync(fullPath).mtimeMs; } catch {}
-    resolvedSnapshots.push({ file: f, artifact, mtime });
-  }
-  resolvedSnapshots.sort((a, b) => a.mtime - b.mtime);
-
-  for (const { file: rf, artifact } of resolvedSnapshots) {
-    try {
-      const content = fs.readFileSync(path.join(dirPath, rf), 'utf-8').trim();
-      const label = artifact.replace(/\.md$/, '').replace(/_/g, ' ');
-      lines.push(`[Snapshot: ${label}]: ${truncate(content, USER_MAX_CHARS)}`);
-      messageCount += 1;
-    } catch {}
-  }
-
-  // Read task.md — the session's current/final goal (after snapshots).
-  // Labeled as CONTEXT rather than USER because Gemini expands task.md with
-  // generated checklists, substeps, and completion markers. The user's
-  // original intent is the title line; the rest is agent-authored.
-  const taskFile = path.join(dirPath, 'task.md');
-  if (fs.existsSync(taskFile)) {
-    const taskContent = fs.readFileSync(taskFile, 'utf-8').trim();
-    lines.push(`[CONTEXT]: ${truncate(taskContent, USER_MAX_CHARS)}`);
-    messageCount += 1;
-  }
-
-  // Read implementation_plan.md — the AI's current/final plan
-  const planFile = path.join(dirPath, 'implementation_plan.md');
-  if (fs.existsSync(planFile)) {
-    const planContent = fs.readFileSync(planFile, 'utf-8').trim();
-    lines.push(`ASSISTANT: ${truncate(planContent, ASSISTANT_MAX_CHARS)}`);
-    messageCount += 1;
-  }
-
-  // Read walkthrough.md — the session's summary/outcome
-  const walkthroughFile = path.join(dirPath, 'walkthrough.md');
-  if (fs.existsSync(walkthroughFile)) {
-    const walkthroughContent = fs.readFileSync(walkthroughFile, 'utf-8').trim();
-    lines.push(`ASSISTANT: ${truncate(walkthroughContent, ASSISTANT_MAX_CHARS)}`);
-    messageCount += 1;
-  }
-
-  return {
-    text: lines.join('\n'),
-    startTimestamp,
-    messageCount,
-  };
-}
-
-function splitIntoSegments(text) {
-  if (text.length <= SEGMENT_THRESHOLD) return [text];
-  const lines = text.split('\n');
-  const segments = [];
-  let current = '';
-
-  for (const line of lines) {
-    if (current.length > 0 && current.length + line.length + 1 > SEGMENT_SIZE) {
-      segments.push(current);
-      current = line;
-    } else {
-      current += (current.length > 0 ? '\n' : '') + line;
-    }
-  }
-  if (current.length > 0) {
-    segments.push(current);
-  }
-  return segments;
-}
-
-// CLI
-if (require.main === module) {
-  const args = process.argv.slice(2);
-  if (args.length < 2) {
-    console.error('Usage: format-session.js <input.jsonl|gemini-brain-dir> <output.txt>');
-    process.exit(1);
-  }
-
-  const inputPath = path.resolve(args[0]);
-  const outputPath = path.resolve(args[1]);
-
-  if (!fs.existsSync(inputPath)) {
-    console.error(`Error: input not found: ${inputPath}`);
-    process.exit(1);
-  }
-
-  try {
-    // Detect Gemini brain entry (directory with task.md) vs JSONL file
-    const isGemini = fs.statSync(inputPath).isDirectory() && fs.existsSync(path.join(inputPath, 'task.md'));
-    const { text, startTimestamp, messageCount } = isGemini
-      ? formatGeminiSession(inputPath)
-      : formatSession(inputPath);
-    fs.mkdirSync(path.dirname(outputPath), { recursive: true });
-
-    const segments = splitIntoSegments(text);
-    const outputFiles = [];
-
-    if (segments.length === 1) {
-      fs.writeFileSync(outputPath, text);
-      outputFiles.push(outputPath);
-    } else {
-      const base = outputPath.replace(/\.txt$/, '');
-      segments.forEach((seg, i) => {
-        const p = `${base}-seg${i + 1}.txt`;
-        fs.writeFileSync(p, seg);
-        outputFiles.push(p);
-      });
-    }
-
-    console.log(JSON.stringify({
-      input: inputPath,
-      start_timestamp: startTimestamp,
-      message_count: messageCount,
-      total_chars: text.length,
-      segments: segments.length,
-      output_files: outputFiles,
-    }, null, 2));
-  } catch (err) {
-    console.error(`Error: ${err.message}`);
-    process.exit(1);
-  }
-}
-
-module.exports = {
-  formatSession, formatGeminiSession, splitIntoSegments,
-  extractMessage: extractMessageClaude,
-  extractMessageCodex, detectFormat, truncate,
-  USER_MAX_CHARS, ASSISTANT_MAX_CHARS, SEGMENT_THRESHOLD, SEGMENT_SIZE,
-};
diff --git a/researchskills-extract/scripts/platform.js b/researchskills-extract/scripts/platform.js
deleted file mode 100644
index caceeca..0000000
--- a/researchskills-extract/scripts/platform.js
+++ /dev/null
@@ -1,195 +0,0 @@
-#!/usr/bin/env node
-/**
- * platform.js
- *
- * Adapter module for spawning AI CLI processes across platforms.
- * Supports Claude Code (`--cc`) and Codex (`--codex`).
- *
- * Usage:
- *   const { parsePlatformFlag, createRunner } = require('./platform');
- *   const platform = parsePlatformFlag();   // reads --cc / --codex from argv
- *   const runner = createRunner(platform);
- *   const { ok, output, error } = await runner.extract(prompt, timeoutMs);
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const { spawn } = require('child_process');
-
-// ---------------------------------------------------------------------------
-// Flag parsing
-// ---------------------------------------------------------------------------
-
-/**
- * Scans process.argv for --cc or --codex, removes the flag, returns the value.
- * Defaults to 'cc' if neither is present (backward compat).
- */
-function parsePlatformFlag() {
-  const idx = process.argv.findIndex(a => a === '--cc' || a === '--codex');
-  if (idx === -1) return 'cc';
-  const flag = process.argv[idx] === '--cc' ? 'cc' : 'codex';
-  process.argv.splice(idx, 1);
-  return flag;
-}
-
-// ---------------------------------------------------------------------------
-// CC (Claude Code) runners
-// ---------------------------------------------------------------------------
-
-function spawnCC(args, prompt, verbose, timeoutMs) {
-  return new Promise((resolve) => {
-    const chunks = [];
-    const proc = spawn('claude', args, { stdio: ['pipe', 'pipe', 'pipe'] });
-
-    proc.stdin.write(prompt);
-    proc.stdin.end();
-
-    proc.stdout.on('data', (d) => {
-      chunks.push(d);
-      if (verbose) process.stderr.write(d);
-    });
-    proc.stderr.on('data', (d) => {
-      if (verbose) process.stderr.write(d);
-    });
-
-    const timer = setTimeout(() => {
-      proc.kill('SIGTERM');
-      resolve({ ok: false, output: Buffer.concat(chunks).toString('utf-8'), error: 'timeout' });
-    }, timeoutMs);
-
-    proc.on('close', (code) => {
-      clearTimeout(timer);
-      const output = Buffer.concat(chunks).toString('utf-8');
-      resolve({ ok: code === 0, output, error: code !== 0 ? `exit ${code}` : null });
-    });
-
-    proc.on('error', (err) => {
-      clearTimeout(timer);
-      resolve({ ok: false, output: '', error: err.message });
-    });
-  });
-}
-
-function createCCRunner() {
-  return {
-    extract(prompt, timeoutMs = 600_000) {
-      return spawnCC(
-        ['-p', '--model', 'sonnet', '--no-session-persistence'],
-        prompt, false, timeoutMs
-      );
-    },
-    clean(prompt, verbose = false, timeoutMs = 900_000) {
-      return spawnCC(
-        ['-p', '--model', 'opus', '--dangerously-skip-permissions', '--allowedTools', 'Read,Edit,Write,Bash'],
-        prompt, verbose, timeoutMs
-      );
-    },
-    score(prompt, verbose = false, timeoutMs = 900_000) {
-      return spawnCC(
-        ['-p', '--model', 'opus', '--dangerously-skip-permissions', '--allowedTools', 'Read,Edit'],
-        prompt, verbose, timeoutMs
-      );
-    },
-    classify(prompt, timeoutMs = 300_000) {
-      return spawnCC(
-        ['-p', '--model', 'sonnet', '--no-session-persistence'],
-        prompt, false, timeoutMs
-      );
-    },
-  };
-}
-
-// ---------------------------------------------------------------------------
-// Codex runners
-// ---------------------------------------------------------------------------
-
-function spawnCodex(args, prompt, verbose, timeoutMs) {
-  return new Promise((resolve) => {
-    const chunks = [];
-    const outputFile = path.join(
-      os.tmpdir(),
-      `codex-last-message-${Date.now()}-${Math.random().toString(36).slice(2)}.txt`
-    );
-
-    const proc = spawn('codex', [...args, '-o', outputFile, '-'], {
-      stdio: ['pipe', 'pipe', 'pipe'],
-    });
-
-    proc.stdin.write(prompt);
-    proc.stdin.end();
-
-    proc.stdout.on('data', (d) => {
-      chunks.push(d);
-      if (verbose) process.stderr.write(d);
-    });
-    proc.stderr.on('data', (d) => {
-      if (verbose) process.stderr.write(d);
-    });
-
-    const timer = setTimeout(() => {
-      proc.kill('SIGTERM');
-      try { fs.unlinkSync(outputFile); } catch {}
-      resolve({ ok: false, output: Buffer.concat(chunks).toString('utf-8'), error: 'timeout' });
-    }, timeoutMs);
-
-    proc.on('close', (code) => {
-      clearTimeout(timer);
-      let output = Buffer.concat(chunks).toString('utf-8');
-      try {
-        const lastMessage = fs.readFileSync(outputFile, 'utf-8').trim();
-        if (lastMessage) output = lastMessage;
-      } catch {}
-      try { fs.unlinkSync(outputFile); } catch {}
-      resolve({ ok: code === 0, output, error: code !== 0 ? `exit ${code}` : null });
-    });
-
-    proc.on('error', (err) => {
-      clearTimeout(timer);
-      try { fs.unlinkSync(outputFile); } catch {}
-      resolve({ ok: false, output: '', error: err.message });
-    });
-  });
-}
-
-function codexBaseArgs(reasoningEffort) {
-  return [
-    '-a', 'never',
-    'exec',
-    '-s', 'danger-full-access',
-    '--ephemeral',
-    '--skip-git-repo-check',
-    '-m', 'gpt-5.4',
-    '-c', `model_reasoning_effort="${reasoningEffort}"`,
-  ];
-}
-
-function createCodexRunner() {
-  return {
-    extract(prompt, timeoutMs = 600_000) {
-      return spawnCodex(codexBaseArgs('medium'), prompt, false, timeoutMs);
-    },
-    clean(prompt, verbose = false, timeoutMs = 900_000) {
-      return spawnCodex(codexBaseArgs('high'), prompt, verbose, timeoutMs);
-    },
-    score(prompt, verbose = false, timeoutMs = 900_000) {
-      return spawnCodex(codexBaseArgs('high'), prompt, verbose, timeoutMs);
-    },
-    classify(prompt, timeoutMs = 300_000) {
-      return spawnCodex(codexBaseArgs('medium'), prompt, false, timeoutMs);
-    },
-  };
-}
-
-// ---------------------------------------------------------------------------
-// Factory
-// ---------------------------------------------------------------------------
-
-function createRunner(platform) {
-  if (platform === 'codex') return createCodexRunner();
-  return createCCRunner();
-}
-
-module.exports = { parsePlatformFlag, createRunner };
diff --git a/researchskills-extract/scripts/postinstall.js b/researchskills-extract/scripts/postinstall.js
deleted file mode 100644
index a91af97..0000000
--- a/researchskills-extract/scripts/postinstall.js
+++ /dev/null
@@ -1,204 +0,0 @@
-#!/usr/bin/env node
-"use strict";
-
-const fs = require("fs");
-const path = require("path");
-const os = require("os");
-
-const SOURCE_CC_COMMAND = path.join(__dirname, "..", "commands", "researchskills-extract.md");
-const SOURCE_CC_CONVERT = path.join(__dirname, "..", "commands", "researchskills-convert.md");
-const SOURCE_CODEX_SKILL = path.join(__dirname, "..", "commands", "SKILL.md");
-const SOURCE_CODEX_CONVERT = path.join(__dirname, "..", "commands", "CONVERT-SKILL.md");
-
-// Helper scripts that must be available at runtime
-const HELPER_SCRIPTS = [
-  "platform.js",
-  "scan-sessions.js",
-  "classify-projects.js",
-  "format-session.js",
-  "extract-skills.js",
-  "validate-skills.js",
-  "clean-skills.js",
-  "score-skills.js",
-  "upload-skills.js",
-  "finalize.js",
-  "store-local.js",
-];
-
-// --- Claude Code ---
-const CC_COMMANDS_DIR = path.join(os.homedir(), ".claude", "commands");
-const CC_COMMAND_TARGET = path.join(CC_COMMANDS_DIR, "researchskills-extract.md");
-const CC_UTILS_DIR = path.join(os.homedir(), ".claude", "utils");
-
-try {
-  fs.mkdirSync(CC_COMMANDS_DIR, { recursive: true });
-  fs.mkdirSync(CC_UTILS_DIR, { recursive: true });
-  fs.copyFileSync(SOURCE_CC_COMMAND, CC_COMMAND_TARGET);
-  const CC_CONVERT_TARGET = path.join(CC_COMMANDS_DIR, "researchskills-convert.md");
-  fs.copyFileSync(SOURCE_CC_CONVERT, CC_CONVERT_TARGET);
-  console.log("✓ Claude Code: /researchskills-extract installed to ~/.claude/commands/");
-  console.log("✓ Claude Code: /researchskills-convert installed to ~/.claude/commands/");
-
-  for (const script of HELPER_SCRIPTS) {
-    const src = path.join(__dirname, script);
-    const dst = path.join(CC_UTILS_DIR, script);
-    if (fs.existsSync(src)) {
-      fs.copyFileSync(src, dst);
-    } else {
-      console.warn(`⚠ Claude Code: ${script} not found in package, skipping`);
-    }
-  }
-  console.log(`✓ Claude Code: ${HELPER_SCRIPTS.length} scripts installed to ~/.claude/utils/`);
-} catch (err) {
-  console.error("⚠ Claude Code: could not install —", err.message);
-}
-
-// --- Codex ---
-const CODEX_SKILL_DIR = path.join(os.homedir(), ".codex", "skills", "researchskills-extract");
-const CODEX_SKILL_TARGET = path.join(CODEX_SKILL_DIR, "SKILL.md");
-const CODEX_SCRIPTS_DIR = path.join(CODEX_SKILL_DIR, "scripts");
-
-try {
-  fs.mkdirSync(CODEX_SKILL_DIR, { recursive: true });
-  fs.mkdirSync(CODEX_SCRIPTS_DIR, { recursive: true });
-  fs.copyFileSync(SOURCE_CODEX_SKILL, CODEX_SKILL_TARGET);
-  console.log("✓ Codex:   /researchskills-extract installed to ~/.codex/skills/researchskills-extract/");
-
-  // --- Codex: researchskills-convert ---
-  const CODEX_CONVERT_DIR = path.join(os.homedir(), ".codex", "skills", "researchskills-convert");
-  fs.mkdirSync(CODEX_CONVERT_DIR, { recursive: true });
-  fs.copyFileSync(SOURCE_CODEX_CONVERT, path.join(CODEX_CONVERT_DIR, "SKILL.md"));
-  console.log("✓ Codex:   /researchskills-convert installed to ~/.codex/skills/researchskills-convert/");
-
-  for (const script of HELPER_SCRIPTS) {
-    const src = path.join(__dirname, script);
-    const dst = path.join(CODEX_SCRIPTS_DIR, script);
-    if (fs.existsSync(src)) {
-      fs.copyFileSync(src, dst);
-    } else {
-      console.warn(`⚠ Codex: ${script} not found in package, skipping`);
-    }
-  }
-  console.log(`✓ Codex:   ${HELPER_SCRIPTS.length} scripts installed to ~/.codex/skills/researchskills-extract/scripts/`);
-} catch (err) {
-  console.error("⚠ Codex: could not install —", err.message);
-}
-
-// --- Clean up legacy extract-knowhow command files ---
-const LEGACY_CC_COMMAND = path.join(os.homedir(), ".claude", "commands", "extract-knowhow.md");
-const LEGACY_CODEX_DIR = path.join(os.homedir(), ".codex", "skills", "extract-knowhow");
-try {
-  if (fs.existsSync(LEGACY_CC_COMMAND)) {
-    fs.unlinkSync(LEGACY_CC_COMMAND);
-    console.log("✓ Removed legacy ~/.claude/commands/extract-knowhow.md");
-  }
-  if (fs.existsSync(LEGACY_CODEX_DIR)) {
-    fs.rmSync(LEGACY_CODEX_DIR, { recursive: true, force: true });
-    console.log("✓ Removed legacy ~/.codex/skills/extract-knowhow/");
-  }
-} catch (err) {
-  console.warn("⚠ Could not clean up legacy files —", err.message);
-}
-
-// --- Migrate legacy cache from @openscientist/extract-knowhow ---
-const OLD_CACHE_ROOT = path.join(os.homedir(), ".openscientist");
-const NEW_CACHE_ROOT = path.join(os.homedir(), ".researchskills");
-
-if (fs.existsSync(OLD_CACHE_ROOT)) {
-  if (!fs.existsSync(NEW_CACHE_ROOT)) {
-    // Simple case: just rename
-    try {
-      fs.renameSync(OLD_CACHE_ROOT, NEW_CACHE_ROOT);
-      console.log("✓ Migrated ~/.openscientist/ → ~/.researchskills/");
-    } catch (err) {
-      console.warn("⚠ Could not migrate ~/.openscientist/ —", err.message);
-      console.warn("  You can manually run: mv ~/.openscientist ~/.researchskills");
-    }
-  } else {
-    // Both exist: recursively merge missing files from old into new
-    function mergeDirs(src, dst) {
-      if (!fs.existsSync(src)) return;
-      fs.mkdirSync(dst, { recursive: true });
-      for (const entry of fs.readdirSync(src)) {
-        const s = path.join(src, entry);
-        const d = path.join(dst, entry);
-        if (fs.statSync(s).isDirectory()) {
-          mergeDirs(s, d); // recurse into subdirs even if they exist
-        } else if (!fs.existsSync(d)) {
-          fs.copyFileSync(s, d);
-        }
-      }
-    }
-    try {
-      for (const sub of ["cache/meta", "cache/skills", "cache/sessions", "skills-fallback"]) {
-        mergeDirs(path.join(OLD_CACHE_ROOT, sub), path.join(NEW_CACHE_ROOT, sub));
-      }
-      console.log("✓ Merged legacy ~/.openscientist/ into ~/.researchskills/");
-    } catch (err) {
-      console.warn("⚠ Could not merge ~/.openscientist/ —", err.message);
-    }
-  }
-}
-
-// --- Handle old package coexistence ---
-// The old @openscientist/extract-knowhow postuninstall deletes the same
-// command files we just installed. Proactively reinstall our files after
-// detecting the old package, so users can safely uninstall it later.
-try {
-  const { execSync } = require("child_process");
-  const out = execSync("npm ls -g @openscientist/extract-knowhow --depth=0 2>/dev/null", { encoding: "utf-8" });
-  if (out.includes("@openscientist/extract-knowhow")) {
-    console.log("\n⚠ Old @openscientist/extract-knowhow is still installed.");
-    console.log("  After uninstalling it, run this to restore command files:");
-    console.log("  npm install -g @scienceintelligence/researchskills-extract\n");
-  }
-} catch (_) { /* not installed or npm ls failed — fine */ }
-
-// --- Cache directory ---
-// Any version change wipes all caches (skills + meta + sessions).
-// Reason: both extraction prompts AND raw-conversation preprocessing can change
-// between versions, so nothing is safe to reuse.
-const CACHE_DIR = path.join(os.homedir(), ".researchskills", "cache");
-const VERSION_FILE = path.join(CACHE_DIR, ".version");
-const CURRENT_VERSION = require(path.join(__dirname, "..", "package.json")).version;
-
-function rmrf(p) {
-  if (!fs.existsSync(p)) return;
-  for (const entry of fs.readdirSync(p)) {
-    const full = path.join(p, entry);
-    if (fs.statSync(full).isDirectory()) rmrf(full);
-    else fs.unlinkSync(full);
-  }
-  fs.rmdirSync(p);
-}
-
-try {
-  fs.mkdirSync(CACHE_DIR, { recursive: true });
-
-  const previousVersion = fs.existsSync(VERSION_FILE)
-    ? fs.readFileSync(VERSION_FILE, "utf-8").trim()
-    : null;
-
-  if (previousVersion && previousVersion !== CURRENT_VERSION) {
-    // Version changed — wipe all processed caches (keep nothing)
-    rmrf(path.join(CACHE_DIR, "skills"));
-    rmrf(path.join(CACHE_DIR, "meta"));
-    rmrf(path.join(CACHE_DIR, "sessions"));
-    console.log(`✓ Cache:       version ${previousVersion} → ${CURRENT_VERSION}, previous cache cleared`);
-  } else if (!previousVersion) {
-    console.log(`✓ Cache:       initialized at version ${CURRENT_VERSION}`);
-  } else {
-    console.log(`✓ Cache:       version ${CURRENT_VERSION} (reusing existing cache)`);
-  }
-
-  fs.mkdirSync(path.join(CACHE_DIR, "meta"), { recursive: true });
-  fs.mkdirSync(path.join(CACHE_DIR, "skills"), { recursive: true });
-  fs.mkdirSync(path.join(CACHE_DIR, "sessions"), { recursive: true });
-  fs.writeFileSync(VERSION_FILE, CURRENT_VERSION);
-} catch (err) {
-  console.error("⚠ Cache: could not prepare —", err.message);
-}
-
-console.log("\n  Usage:");
-console.log("    /researchskills-extract  (Claude Code) or $researchskills-extract  (Codex) — extract from history");
-console.log("    /researchskills-convert  (Claude Code) or $researchskills-convert  (Codex) — convert existing skills");
diff --git a/researchskills-extract/scripts/postuninstall.js b/researchskills-extract/scripts/postuninstall.js
deleted file mode 100644
index dd3133e..0000000
--- a/researchskills-extract/scripts/postuninstall.js
+++ /dev/null
@@ -1,83 +0,0 @@
-#!/usr/bin/env node
-"use strict";
-
-const fs = require("fs");
-const path = require("path");
-const os = require("os");
-
-const HELPER_SCRIPTS = [
-  "platform.js",
-  "scan-sessions.js",
-  "classify-projects.js",
-  "format-session.js",
-  "extract-skills.js",
-  "validate-skills.js",
-  "clean-skills.js",
-  "score-skills.js",
-  "upload-skills.js",
-  "finalize.js",
-  "store-local.js",
-];
-
-// --- Claude Code ---
-const CC_COMMAND_TARGET = path.join(os.homedir(), ".claude", "commands", "researchskills-extract.md");
-const CC_CONVERT_TARGET = path.join(os.homedir(), ".claude", "commands", "researchskills-convert.md");
-const CC_UTILS_DIR = path.join(os.homedir(), ".claude", "utils");
-
-try {
-  if (fs.existsSync(CC_COMMAND_TARGET)) {
-    fs.unlinkSync(CC_COMMAND_TARGET);
-    console.log("✓ Claude Code: /researchskills-extract removed");
-  }
-  if (fs.existsSync(CC_CONVERT_TARGET)) {
-    fs.unlinkSync(CC_CONVERT_TARGET);
-    console.log("✓ Claude Code: /researchskills-convert removed");
-  }
-  for (const script of HELPER_SCRIPTS) {
-    const p = path.join(CC_UTILS_DIR, script);
-    if (fs.existsSync(p)) fs.unlinkSync(p);
-  }
-  console.log("✓ Claude Code: helper scripts removed");
-} catch (err) {
-  // ignore
-}
-
-// --- Codex ---
-const CODEX_SKILL_DIR = path.join(os.homedir(), ".codex", "skills", "researchskills-extract");
-const CODEX_SKILL_TARGET = path.join(CODEX_SKILL_DIR, "SKILL.md");
-const CODEX_SCRIPTS_DIR = path.join(CODEX_SKILL_DIR, "scripts");
-
-try {
-  if (fs.existsSync(CODEX_SKILL_TARGET)) {
-    fs.unlinkSync(CODEX_SKILL_TARGET);
-    console.log("✓ Codex: /researchskills-extract SKILL.md removed");
-  }
-  for (const script of HELPER_SCRIPTS) {
-    const p = path.join(CODEX_SCRIPTS_DIR, script);
-    if (fs.existsSync(p)) fs.unlinkSync(p);
-  }
-  // Remove dirs if empty
-  try {
-    const remaining = fs.readdirSync(CODEX_SCRIPTS_DIR);
-    if (remaining.length === 0) fs.rmdirSync(CODEX_SCRIPTS_DIR);
-    const skillRemaining = fs.readdirSync(CODEX_SKILL_DIR);
-    if (skillRemaining.length === 0) fs.rmdirSync(CODEX_SKILL_DIR);
-  } catch (_) { /* best effort */ }
-  console.log("✓ Codex: helper scripts removed");
-} catch (err) {
-  // ignore
-}
-
-// --- Codex: researchskills-convert ---
-const CODEX_CONVERT_DIR = path.join(os.homedir(), ".codex", "skills", "researchskills-convert");
-try {
-  if (fs.existsSync(CODEX_CONVERT_DIR)) {
-    fs.rmSync(CODEX_CONVERT_DIR, { recursive: true, force: true });
-    console.log("✓ Codex: /researchskills-convert removed");
-  }
-} catch (err) {
-  // ignore
-}
-
-// Note: cache directory ~/.researchskills/cache/ is intentionally preserved,
-// so reinstalling retains previously extracted subtrees.
diff --git a/researchskills-extract/scripts/scan-sessions.js b/researchskills-extract/scripts/scan-sessions.js
deleted file mode 100644
index 3a65afd..0000000
--- a/researchskills-extract/scripts/scan-sessions.js
+++ /dev/null
@@ -1,598 +0,0 @@
-#!/usr/bin/env node
-/**
- * scan-sessions.js
- *
- * Stage 1 + 2 of /researchskills-extract, as a single deterministic script.
- *
- * Discovers Claude Code, Codex, and Gemini brain session files, extracts per-session
- * metadata, filters out garbage (too-small, too-short, sub-agent, duplicate),
- * groups by project path, and emits a work list that the AI phase iterates
- * over.
- *
- * Output shape (work-list.json):
- *   {
- *     generated_at: ISO,
- *     totals: { discovered, accepted },
- *     skipped: { tooSmall, tooShort, subAgent, duplicate, unreadable },
- *     sessions: [
- *       { session_id, source, file_path, file_size, project_path,
- *         first_prompt, user_message_count, duration_minutes,
- *         start_timestamp }
- *     ],
- *     projects: { "<project_path>": [session_id, ...] }
- *   }
- *
- * Per-session metadata is cached at ~/.researchskills/cache/meta/<id>.json
- * keyed by file_size, so unchanged sessions are not re-parsed.
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-
-const META_CACHE_DIR = path.join(os.homedir(), '.researchskills', 'cache', 'meta');
-const MIN_FILE_SIZE = 500;
-const MIN_USER_MESSAGES = 2;
-const MIN_DURATION_MINUTES = 1;
-const HEAD_LINES = 50;
-const TAIL_LINES = 20;
-const SUBAGENT_MARKERS = [
-  'RESPOND WITH ONLY A VALID JSON OBJECT',
-  'record_facets',
-];
-
-function walk(dir, matcher, out = []) {
-  let entries;
-  try {
-    entries = fs.readdirSync(dir, { withFileTypes: true });
-  } catch (err) {
-    return out;
-  }
-  for (const e of entries) {
-    const full = path.join(dir, e.name);
-    if (e.isDirectory()) walk(full, matcher, out);
-    else if (e.isFile() && matcher(full)) out.push(full);
-  }
-  return out;
-}
-
-function discoverClaude() {
-  const root = path.join(os.homedir(), '.claude', 'projects');
-  const sep = path.sep;
-  return walk(
-    root,
-    (p) => p.endsWith('.jsonl') && !p.includes(`${sep}subagents${sep}`)
-  );
-}
-
-function discoverCodex() {
-  const archived = walk(
-    path.join(os.homedir(), '.codex', 'archived_sessions'),
-    (p) => /rollout-.*\.jsonl$/.test(p)
-  );
-  const sessions = walk(
-    path.join(os.homedir(), '.codex', 'sessions'),
-    (p) => p.endsWith('.jsonl')
-  );
-  return [...archived, ...sessions];
-}
-
-/**
- * Discover Gemini brain entries at ~/.gemini/antigravity/brain/<uuid>/.
- * Each non-empty UUID directory is treated as one session.
- * Returns an array of directory paths (not individual files).
- */
-function discoverGemini() {
-  const root = path.join(os.homedir(), '.gemini', 'antigravity', 'brain');
-  const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/;
-  const results = [];
-  let entries;
-  try {
-    entries = fs.readdirSync(root, { withFileTypes: true });
-  } catch (err) {
-    return results;
-  }
-  for (const e of entries) {
-    if (!e.isDirectory() || !UUID_RE.test(e.name)) continue;
-    const dir = path.join(root, e.name);
-    const taskFile = path.join(dir, 'task.md');
-    if (!fs.existsSync(taskFile)) continue;
-    results.push(dir);
-  }
-  return results;
-}
-
-function extractSessionId(filePath, source) {
-  if (source === 'gemini') {
-    // filePath is the brain entry directory; its basename is the UUID
-    return path.basename(filePath);
-  }
-  const base = path.basename(filePath, '.jsonl');
-  if (source === 'codex') {
-    const m = base.match(/([0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})/);
-    if (m) return m[1];
-  }
-  return base;
-}
-
-function claudeProjectDirName(filePath) {
-  return path.basename(path.dirname(filePath));
-}
-
-function parseLine(line) {
-  try {
-    return JSON.parse(line);
-  } catch (err) {
-    return null;
-  }
-}
-
-const BOILERPLATE_PREFIXES = [
-  '<environment_context>',
-  '<user_instructions>',
-  '<system-reminder>',
-  '# AGENTS.md',
-  '<INSTRUCTIONS>',
-  '<command-name>',
-  '<local-command',
-  '[Tool Result',
-  '<turn_aborted>',
-];
-
-const IDE_CONTEXT_RE = /^# Context from my IDE.*?\n(?:(?:##[^\n]*|  ?- [^\n]*|-[^\n]*|\s*)\n)*\s*/s;
-
-function firstPromptText(entry) {
-  // Claude Code: { type, message: { role, content } }
-  // Codex:   { type: "event_msg", payload: { type: "user_message", message } }
-  const msg = entry.message || entry.payload || entry;
-  const role =
-    msg.role ||
-    entry.role ||
-    (entry.type === 'user' ? 'user' : entry.type === 'assistant' ? 'assistant' : null);
-
-  // Codex user_message
-  if (entry.type === 'event_msg' && entry.payload && entry.payload.type === 'user_message') {
-    const text = (entry.payload.message || '').trim();
-    if (!text) return null;
-    if (BOILERPLATE_PREFIXES.some((b) => text.startsWith(b))) return null;
-    if (text.includes('RESPOND WITH ONLY A VALID JSON OBJECT')) return null;
-    return text.replace(IDE_CONTEXT_RE, '').trim() || null;
-  }
-
-  if (role !== 'user') return null;
-
-  const content = msg.content;
-  let text = '';
-  if (typeof content === 'string') {
-    text = content;
-  } else if (Array.isArray(content)) {
-    for (const p of content) {
-      if (!p) continue;
-      if (typeof p === 'string') text += p;
-      else if ((p.type === 'text' || p.type === 'input_text') && p.text) text += p.text;
-    }
-  }
-  text = text.trim();
-  if (!text) return null;
-  if (BOILERPLATE_PREFIXES.some((b) => text.startsWith(b))) return null;
-  if (text.includes('RESPOND WITH ONLY A VALID JSON OBJECT')) return null;
-
-  text = text.replace(IDE_CONTEXT_RE, '').trim();
-  if (!text) return null;
-
-  return text;
-}
-
-function timestampOf(entry) {
-  const msg = entry.message || entry.payload || {};
-  return entry.timestamp || entry.ts || msg.timestamp || null;
-}
-
-function extractMeta(filePath) {
-  const stats = fs.statSync(filePath);
-  const content = fs.readFileSync(filePath, 'utf-8');
-  const lines = content.split('\n').filter((l) => l.trim());
-
-  const head = lines.slice(0, HEAD_LINES);
-  const tail = lines.slice(-TAIL_LINES);
-
-  const firstFive = lines.slice(0, 5).join('\n');
-  const isSubAgent = SUBAGENT_MARKERS.some((m) => firstFive.includes(m));
-
-  let firstPrompt = null;
-  let cwd = null;
-  let firstTs = null;
-
-  for (const line of head) {
-    const entry = parseLine(line);
-    if (!entry) continue;
-    if (!cwd) {
-      cwd =
-        entry.cwd ||
-        (entry.payload && entry.payload.cwd) ||
-        (entry.message && entry.message.cwd) ||
-        null;
-    }
-    if (!firstPrompt) {
-      const t = firstPromptText(entry);
-      if (t) firstPrompt = t.substring(0, 500);
-    }
-    if (!firstTs) {
-      const ts = timestampOf(entry);
-      if (ts) firstTs = ts;
-    }
-    if (firstPrompt && firstTs && cwd) break;
-  }
-
-  let lastTs = null;
-  for (let i = tail.length - 1; i >= 0; i--) {
-    const entry = parseLine(tail[i]);
-    if (!entry) continue;
-    const ts = timestampOf(entry);
-    if (ts) {
-      lastTs = ts;
-      break;
-    }
-  }
-
-  const userLineIndices = [];
-  for (let li = 0; li < lines.length; li++) {
-    if (lines[li].includes('"role":"user"') ||
-        lines[li].includes('"type":"user"') ||
-        lines[li].includes('"type":"user_message"')) {
-      userLineIndices.push(li);
-    }
-  }
-  const userCount = userLineIndices.length;
-
-  const SAMPLE_COUNT = 5;
-  const sampledPrompts = [];
-  if (userLineIndices.length > 0) {
-    const pickIndices = new Set();
-    for (let s = 0; s < SAMPLE_COUNT; s++) {
-      const idx = Math.min(
-        Math.round(s * (userLineIndices.length - 1) / Math.max(SAMPLE_COUNT - 1, 1)),
-        userLineIndices.length - 1
-      );
-      pickIndices.add(userLineIndices[idx]);
-    }
-    for (const li of pickIndices) {
-      const entry = parseLine(lines[li]);
-      if (!entry) continue;
-      const t = firstPromptText(entry);
-      if (t) sampledPrompts.push(t.substring(0, 300));
-    }
-  }
-
-  let durationMinutes = 0;
-  if (firstTs && lastTs) {
-    const dt = new Date(lastTs).getTime() - new Date(firstTs).getTime();
-    if (!Number.isNaN(dt) && dt > 0) {
-      durationMinutes = Math.round(dt / 60000);
-    }
-  }
-
-  return {
-    file_size: stats.size,
-    is_sub_agent: isSubAgent,
-    first_prompt: firstPrompt,
-    sampled_prompts: sampledPrompts,
-    user_message_count: userCount,
-    cwd,
-    start_timestamp: firstTs,
-    end_timestamp: lastTs,
-    duration_minutes: durationMinutes,
-  };
-}
-
-/**
- * Extract metadata from a Gemini brain entry directory.
- * Brain entries contain task.md, implementation_plan.md, walkthrough.md,
- * plus .metadata.json sidecars and .resolved.N historical snapshots.
- */
-function extractMetaGemini(dirPath) {
-  const taskFile = path.join(dirPath, 'task.md');
-  const taskContent = fs.readFileSync(taskFile, 'utf-8');
-  const titleMatch = taskContent.match(/^#\s+(.+)/m);
-  const firstPrompt = titleMatch ? titleMatch[1].trim() : null;
-
-  const allFiles = fs.readdirSync(dirPath);
-
-  // Compute combined size of all files (matches the cache key in scan())
-  let totalSize = 0;
-  for (const f of allFiles) {
-    try { totalSize += fs.statSync(path.join(dirPath, f)).size; } catch {}
-  }
-
-  // Gemini brain entries don't reliably distinguish user-initiated turns
-  // from agent-generated artifact updates. Count distinct metadata
-  // timestamps separated by >60s as separate interactions — timestamps
-  // within the same minute are assumed to be from a single agent run.
-  const allMetaTimestamps = [];
-  for (const f of allFiles) {
-    if (!f.endsWith('.metadata.json')) continue;
-    try {
-      const meta = JSON.parse(fs.readFileSync(path.join(dirPath, f), 'utf-8'));
-      if (meta.updatedAt) allMetaTimestamps.push(new Date(meta.updatedAt).getTime());
-    } catch {}
-  }
-  allMetaTimestamps.sort((a, b) => a - b);
-  let userMessageCount = allMetaTimestamps.length > 0 ? 1 : 0;
-  for (let i = 1; i < allMetaTimestamps.length; i++) {
-    if (allMetaTimestamps[i] - allMetaTimestamps[i - 1] > 60000) {
-      userMessageCount++;
-    }
-  }
-
-  // Extract timestamps from metadata sidecars
-  let startTimestamp = null;
-  let endTimestamp = null;
-  for (const f of allFiles) {
-    if (!f.endsWith('.metadata.json')) continue;
-    try {
-      const meta = JSON.parse(fs.readFileSync(path.join(dirPath, f), 'utf-8'));
-      const ts = meta.updatedAt || null;
-      if (ts) {
-        if (!startTimestamp || ts < startTimestamp) startTimestamp = ts;
-        if (!endTimestamp || ts > endTimestamp) endTimestamp = ts;
-      }
-    } catch {}
-  }
-
-  // Extract project path from file:// links in plan, task, and resolved snapshots
-  let cwd = null;
-  const planFile = path.join(dirPath, 'implementation_plan.md');
-  const searchFiles = [planFile, taskFile];
-  // Also search resolved snapshots which may contain file:// links
-  for (const f of allFiles) {
-    if (/\.resolved(\.\d+)?$/.test(f)) {
-      searchFiles.push(path.join(dirPath, f));
-    }
-  }
-  for (const sf of searchFiles) {
-    if (cwd) break;
-    try {
-      const content = fs.readFileSync(sf, 'utf-8');
-      const fileMatch = content.match(/file:\/\/\/([\w/._-]+\/[\w._-]+)/);
-      if (fileMatch) {
-        const filePath = '/' + fileMatch[1];
-        // Walk up to find a likely project root (dir with package.json, Cargo.toml, etc.)
-        const parts = filePath.split('/');
-        for (let i = parts.length - 1; i >= 2; i--) {
-          const candidate = parts.slice(0, i).join('/');
-          try {
-            const entries = fs.readdirSync(candidate);
-            if (entries.some((e) => ['package.json', 'Cargo.toml', 'go.mod', 'pyproject.toml', '.git'].includes(e))) {
-              cwd = candidate;
-              break;
-            }
-          } catch {}
-        }
-        // If the linked directory no longer exists (e.g. deleted worktree),
-        // leave cwd unset. scan() will fall back to gemini/<uuid> which
-        // avoids splitting one repo across file-level pseudo-projects.
-      }
-    } catch {}
-  }
-
-  let durationMinutes = 0;
-  if (startTimestamp && endTimestamp) {
-    const dt = new Date(endTimestamp).getTime() - new Date(startTimestamp).getTime();
-    if (!Number.isNaN(dt) && dt > 0) {
-      durationMinutes = Math.round(dt / 60000);
-    }
-  }
-
-  // Fallback: derive duration from artifact mtimes ONLY when metadata has
-  // no usable span (start==end or missing). Do not fall back when metadata
-  // yields a real sub-minute span (rounded to 0), as old resolved snapshots
-  // can be days older and would inflate the duration.
-  const hasMetadataSpan = startTimestamp && endTimestamp && startTimestamp !== endTimestamp;
-  if (durationMinutes === 0 && !hasMetadataSpan) {
-    const artifactMtimes = [];
-    for (const f of allFiles) {
-      if (/\.resolved\.\d+$/.test(f) || f === 'task.md') {
-        try { artifactMtimes.push(fs.statSync(path.join(dirPath, f)).mtimeMs); } catch {}
-      }
-    }
-    if (artifactMtimes.length >= 2) {
-      const span = Math.max(...artifactMtimes) - Math.min(...artifactMtimes);
-      if (span > 0) durationMinutes = Math.round(span / 60000);
-    }
-  }
-
-  return {
-    file_size: totalSize,
-    is_sub_agent: false,
-    first_prompt: firstPrompt ? firstPrompt.substring(0, 500) : null,
-    sampled_prompts: [],
-    user_message_count: userMessageCount,
-    cwd,
-    start_timestamp: startTimestamp,
-    end_timestamp: endTimestamp,
-    duration_minutes: durationMinutes,
-  };
-}
-
-function metaCachePath(sessionId) {
-  return path.join(META_CACHE_DIR, `${sessionId}.json`);
-}
-
-function loadCachedMeta(sessionId, fileSize) {
-  const p = metaCachePath(sessionId);
-  if (!fs.existsSync(p)) return null;
-  try {
-    const cached = JSON.parse(fs.readFileSync(p, 'utf-8'));
-    if (cached && cached.file_size === fileSize) return cached;
-  } catch (err) {
-    // fall through to re-parse
-  }
-  return null;
-}
-
-function saveCachedMeta(sessionId, meta) {
-  fs.mkdirSync(META_CACHE_DIR, { recursive: true });
-  fs.writeFileSync(metaCachePath(sessionId), JSON.stringify(meta, null, 2) + '\n');
-}
-
-function scan() {
-  const candidates = [
-    ...discoverClaude().map((f) => ({ file: f, source: 'claude' })),
-    ...discoverCodex().map((f) => ({ file: f, source: 'codex' })),
-    ...discoverGemini().map((f) => ({ file: f, source: 'gemini' })),
-  ];
-
-  const skipped = {
-    tooSmall: 0,
-    tooShort: 0,
-    subAgent: 0,
-    duplicate: 0,
-    unreadable: 0,
-  };
-  const accepted = [];
-  const byFingerprint = new Map();
-
-  for (const { file, source } of candidates) {
-    // For Gemini, `file` is a directory; sum all file sizes so the cache
-    // key invalidates when any artifact changes (resolved, metadata, etc.)
-    let fileSize;
-    try {
-      if (source === 'gemini') {
-        fileSize = 0;
-        for (const f of fs.readdirSync(file)) {
-          try { fileSize += fs.statSync(path.join(file, f)).size; } catch {}
-        }
-      } else {
-        fileSize = fs.statSync(file).size;
-      }
-    } catch (err) {
-      skipped.unreadable += 1;
-      continue;
-    }
-    if (fileSize < MIN_FILE_SIZE) {
-      skipped.tooSmall += 1;
-      continue;
-    }
-
-    const sessionId = extractSessionId(file, source);
-    let meta = loadCachedMeta(sessionId, fileSize);
-    if (!meta) {
-      try {
-        meta = source === 'gemini' ? extractMetaGemini(file) : extractMeta(file);
-      } catch (err) {
-        skipped.unreadable += 1;
-        continue;
-      }
-      meta.session_id = sessionId;
-      meta.source = source;
-      meta.file_path = file;
-      saveCachedMeta(sessionId, meta);
-    }
-
-    if (meta.is_sub_agent) {
-      skipped.subAgent += 1;
-      continue;
-    }
-    if ((meta.user_message_count || 0) < MIN_USER_MESSAGES) {
-      skipped.tooShort += 1;
-      continue;
-    }
-    if ((meta.duration_minutes || 0) < MIN_DURATION_MINUTES) {
-      skipped.tooShort += 1;
-      continue;
-    }
-
-    let projectPath = meta.cwd;
-    if (!projectPath) {
-      if (source === 'claude') {
-        projectPath = claudeProjectDirName(file);
-      } else if (source === 'gemini') {
-        // Use gemini/<uuid> to keep link-less entries separate
-        projectPath = `gemini/${path.basename(file)}`;
-      } else {
-        projectPath = path.basename(path.dirname(file));
-      }
-    }
-
-    const record = {
-      session_id: sessionId,
-      source,
-      file_path: file,
-      file_size: meta.file_size,
-      project_path: projectPath,
-      first_prompt: meta.first_prompt,
-      sampled_prompts: meta.sampled_prompts || [],
-      user_message_count: meta.user_message_count,
-      duration_minutes: meta.duration_minutes,
-      start_timestamp: meta.start_timestamp,
-    };
-
-    const fp = `${projectPath}::${meta.first_prompt || ''}::${meta.user_message_count || 0}`;
-    const existing = byFingerprint.get(fp);
-    if (existing) {
-      skipped.duplicate += 1;
-      if ((record.user_message_count || 0) > (existing.user_message_count || 0)) {
-        const idx = accepted.indexOf(existing);
-        if (idx !== -1) accepted[idx] = record;
-        byFingerprint.set(fp, record);
-      }
-      continue;
-    }
-    accepted.push(record);
-    byFingerprint.set(fp, record);
-  }
-
-  accepted.sort((a, b) => {
-    const at = a.start_timestamp || '';
-    const bt = b.start_timestamp || '';
-    return at.localeCompare(bt);
-  });
-
-  const projects = {};
-  for (const s of accepted) {
-    if (!projects[s.project_path]) projects[s.project_path] = [];
-    projects[s.project_path].push(s.session_id);
-  }
-
-  return {
-    generated_at: new Date().toISOString(),
-    totals: {
-      discovered: candidates.length,
-      accepted: accepted.length,
-    },
-    skipped,
-    sessions: accepted,
-    projects,
-  };
-}
-
-// CLI
-if (require.main === module) {
-  const args = process.argv.slice(2);
-  const outIdx = args.indexOf('--out');
-  const outPath =
-    outIdx !== -1 && args[outIdx + 1]
-      ? path.resolve(args[outIdx + 1])
-      : path.join(os.homedir(), '.researchskills', 'cache', 'work-list.json');
-
-  try {
-    const result = scan();
-    fs.mkdirSync(path.dirname(outPath), { recursive: true });
-    fs.writeFileSync(outPath, JSON.stringify(result, null, 2) + '\n');
-    console.log(
-      `✓ Scanned ${result.totals.discovered} files → ${result.totals.accepted} accepted`
-    );
-    console.log(
-      `  Skipped: tooSmall=${result.skipped.tooSmall} tooShort=${result.skipped.tooShort} subAgent=${result.skipped.subAgent} duplicate=${result.skipped.duplicate} unreadable=${result.skipped.unreadable}`
-    );
-    console.log(`  Projects: ${Object.keys(result.projects).length}`);
-    console.log(`  Work list: ${outPath}`);
-  } catch (err) {
-    console.error(`Error: ${err.message}`);
-    process.exit(1);
-  }
-}
-
-module.exports = { scan, extractMeta };
diff --git a/researchskills-extract/scripts/score-skills.js b/researchskills-extract/scripts/score-skills.js
deleted file mode 100644
index 2560413..0000000
--- a/researchskills-extract/scripts/score-skills.js
+++ /dev/null
@@ -1,222 +0,0 @@
-#!/usr/bin/env node
-/**
- * score-skills.js
- *
- * Stage 5 of /researchskills-extract: score cleaned skills.
- * Spawns an AI instance (via platform.js) that reads
- * each skill file and writes review_scores into the YAML frontmatter.
- *
- * Three scoring dimensions (0-5 each):
- *   - procedural: does it provide decision frameworks AI doesn't know?
- *   - semantic:   does it provide facts/beliefs AI doesn't have?
- *   - episodic:   does it provide concrete research experiences AI can reference?
- *
- * Usage:
- *   score-skills.js --session-ids id1,id2,... [--verbose]
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const { parsePlatformFlag, createRunner } = require('./platform');
-
-const CACHE_DIR = path.join(os.homedir(), '.researchskills', 'cache', 'skills');
-
-// ---------------------------------------------------------------------------
-// CLI
-// ---------------------------------------------------------------------------
-
-function parseArgs() {
-  const args = process.argv.slice(2);
-  const opts = { sessionIds: null, verbose: false };
-  for (let i = 0; i < args.length; i++) {
-    switch (args[i]) {
-      case '--session-ids': opts.sessionIds = args[++i].split(','); break;
-      case '--verbose':     opts.verbose = true; break;
-    }
-  }
-  return opts;
-}
-
-// ---------------------------------------------------------------------------
-// Collect skill files
-// ---------------------------------------------------------------------------
-
-function collectSkillFiles(sessionIds) {
-  const files = [];
-  for (const sid of sessionIds) {
-    const dir = path.join(CACHE_DIR, sid);
-    if (!fs.existsSync(dir)) continue;
-    for (const f of fs.readdirSync(dir)) {
-      if (f.endsWith('.md')) {
-        files.push(path.join(dir, f));
-      }
-    }
-  }
-  return files;
-}
-
-// ---------------------------------------------------------------------------
-// Build prompt
-// ---------------------------------------------------------------------------
-
-function buildPrompt(files) {
-  const fileList = files.map(f => f).join('\n');
-
-  return `You are a research knowledge value assessor. You have ${files.length} cleaned research skill files to score.
-
-These skills have already passed quality review (engineering content removed, PII fixed, duplicates merged). Your job is to assess the VALUE each skill brings to the most capable AI systems.
-
-## Skill file paths
-
-<files>
-${fileList}
-</files>
-
-## Instructions
-
-Read each file using the Read tool. For each skill, assign three independent scores (0-5), then use the Edit tool to insert them into the file's YAML frontmatter.
-
-Insert the following block just before the closing \`---\` of the frontmatter:
-
-\`\`\`
-review_scores:
-  procedural: X
-  semantic: Y
-  episodic: Z
-\`\`\`
-
-If the file already has a \`review_scores\` block, replace it with the new scores.
-
-## Three Scoring Dimensions
-
-### procedural — Procedural memory value
-**Question**: Does this skill provide research decision frameworks or principles that the strongest AI doesn't already know?
-
-| Score | Meaning |
-|-------|---------|
-| 0 | AI can fully derive this decision logic on its own, no external prompt needed |
-| 1 | AI would likely think of this, but might not prioritize this path |
-| 2 | AI could figure out parts of this, but would miss key exclusion conditions or failure handling |
-| 3 | AI is unlikely to independently produce this decision framework, but would recognize it as sound |
-| 4 | AI would take the wrong path when facing this research impasse; this skill directly corrects its search direction |
-| 5 | AI would confidently take the wrong path; this skill corrects a confident-but-wrong decision pattern |
-
-**Focus**: Is the trigger condition specific? Are rejected alternatives explicitly stated with reasons? Is failure recovery covered?
-
-### semantic — Declarative memory value
-**Question**: Does this skill provide knowledge or beliefs that the strongest AI doesn't have?
-
-| Score | Meaning |
-|-------|---------|
-| 0 | Textbook-level knowledge, any LLM knows this |
-| 1 | Public but obscure knowledge, model might know but uncertain |
-| 2 | Relatively new knowledge, may have appeared after model training cutoff |
-| 3 | Model very likely doesn't know this specific fact (e.g., undocumented tool behavior) |
-| 4 | Model holds an incorrect belief here, would give a confident but wrong answer |
-| 5 | Non-public, lab-level knowledge that cannot exist in training data |
-
-**Focus**: Is there concrete evidence? Is it frontier / non-public / correction type?
-
-### episodic — Episodic memory value
-**Question**: Does this skill provide concrete research experiences that AI can reference?
-
-| Score | Meaning |
-|-------|---------|
-| 0 | Pure abstract advice, no concrete situation |
-| 1 | Has situational description but very generic ("in an experiment...") |
-| 2 | Has specific situation and action, but outcome/lesson unclear |
-| 3 | Complete situation-action-outcome chain, AI can reference in similar situations |
-| 4 | Contains counter-intuitive turning point (expected A, observed B), AI can directly reuse when encountering similar anomalies |
-| 5 | Highly specific failure/adaptation/anomaly experience with clear retrieval cues that can auto-trigger in similar situations |
-
-**Focus**: Is there a concrete situation → action → outcome chain? Is the lesson transferable?
-
-## Key Principles
-
-- **Three dimensions are independent**: A procedural skill can simultaneously score high on semantic (carries facts AI doesn't know) and episodic (grounded in a specific research episode)
-- **Core criterion**: Without this skill, would the strongest frontier AI perform worse in the corresponding dimension?
-- **memory_type vs review_scores**: \`memory_type\` is structural classification (format). \`review_scores\` is value assessment (what it actually contributes).
-
-## Output
-
-After scoring ALL files, output this exact line as the very last line of your response:
-
-SCORE_SUMMARY: scored=<N> avg_procedural=<X.X> avg_semantic=<X.X> avg_episodic=<X.X>
-
-Now begin. Read each file and score them systematically.`;
-}
-
-// ---------------------------------------------------------------------------
-// Parse summary from output
-// ---------------------------------------------------------------------------
-
-function parseSummary(output) {
-  const match = output.match(
-    /SCORE_SUMMARY:\s*scored=(\d+)\s+avg_procedural=([\d.]+)\s+avg_semantic=([\d.]+)\s+avg_episodic=([\d.]+)/
-  );
-  if (!match) return null;
-  return {
-    scored: parseInt(match[1], 10),
-    avg_procedural: parseFloat(match[2]),
-    avg_semantic: parseFloat(match[3]),
-    avg_episodic: parseFloat(match[4]),
-  };
-}
-
-// ---------------------------------------------------------------------------
-// Main
-// ---------------------------------------------------------------------------
-
-async function main() {
-  const platform = parsePlatformFlag();
-  const runner = createRunner(platform);
-  const opts = parseArgs();
-
-  if (!opts.sessionIds || opts.sessionIds.length === 0) {
-    console.error('Usage: score-skills.js --session-ids id1,id2,... [--cc|--codex] [--verbose]');
-    process.exit(1);
-  }
-
-  const files = collectSkillFiles(opts.sessionIds);
-  console.log(`\nScore: ${files.length} skill files across ${opts.sessionIds.length} sessions`);
-
-  if (files.length === 0) {
-    console.log('No skill files to score. Done.');
-    process.exit(0);
-  }
-
-  console.log('Spawning AI for scoring...\n');
-  const prompt = buildPrompt(files);
-  const { ok, output, error } = await runner.score(prompt, opts.verbose);
-
-  if (!ok) {
-    console.error(`\nScoring failed: ${error}`);
-    process.exit(1);
-  }
-
-  const summary = parseSummary(output);
-  if (summary) {
-    console.log(`
-═══════════════════════════════════════════════
-  Score Complete
-═══════════════════════════════════════════════
-  Scored:         ${summary.scored} skills
-  Avg procedural: ${summary.avg_procedural.toFixed(1)}
-  Avg semantic:   ${summary.avg_semantic.toFixed(1)}
-  Avg episodic:   ${summary.avg_episodic.toFixed(1)}
-═══════════════════════════════════════════════`);
-  } else {
-    console.log('\nCompleted but no SCORE_SUMMARY found in output.');
-    if (!opts.verbose) {
-      console.log('Re-run with --verbose to see full output.');
-    }
-  }
-}
-
-main().catch(err => {
-  console.error('Fatal:', err);
-  process.exit(1);
-});
diff --git a/researchskills-extract/scripts/store-local.js b/researchskills-extract/scripts/store-local.js
deleted file mode 100644
index 4bd9fc8..0000000
--- a/researchskills-extract/scripts/store-local.js
+++ /dev/null
@@ -1,231 +0,0 @@
-#!/usr/bin/env node
-/**
- * store-local.js
- *
- * Install extracted research skills into the user's local AI coding tool
- * configuration so they are available as commands/skills in future sessions.
- *
- * Targets:
- *   - Claude Code: ~/.claude/commands/researchskills/<slug>.md
- *   - Codex:       ~/.codex/skills/researchskills-<slug>/SKILL.md
- *
- * Skill slugs are derived from the YAML frontmatter `name` field, not from
- * cache temp filenames. Deduplication uses the slug, so re-extractions that
- * produce the same skill name will overwrite cleanly.
- *
- * Usage:
- *   store-local.js --target claude|codex|both --session-ids id1,id2,...
- *
- * Reads validated skills from ~/.researchskills/cache/skills/<session_id>/
- * and copies them to the chosen target(s).
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const crypto = require('crypto');
-
-const CACHE_DIR = path.join(os.homedir(), '.researchskills', 'cache', 'skills');
-
-/**
- * Extract a field from YAML frontmatter.
- * Returns null if not found.
- */
-function extractField(content, field) {
-  const match = content.match(/^---\s*\n([\s\S]*?)\n---/);
-  if (!match) return null;
-  const fieldMatch = match[1].match(new RegExp(`^${field}:\\s*["']?(.+?)["']?\\s*$`, 'm'));
-  return fieldMatch ? fieldMatch[1].trim() : null;
-}
-
-/**
- * Turn a skill name into a filesystem-safe slug.
- */
-function slugify(name) {
-  return String(name)
-    .toLowerCase()
-    .replace(/[^a-z0-9]+/g, '-')
-    .replace(/^-+|-+$/g, '')
-    .substring(0, 80) || 'unnamed-skill';
-}
-
-/**
- * Hash the contributor field for de-identification (matches validate-skills.js).
- */
-function hashContributor(content) {
-  const match = content.match(/^contributor:\s*(.+)$/m);
-  if (!match) return content;
-  const raw = match[1].trim().replace(/^["']|["']$/g, '');
-  if (raw.startsWith('anon-')) return content; // already hashed
-  const hash = crypto.createHash('sha256').update(raw).digest('hex').substring(0, 8);
-  return content.replace(/^contributor:\s*.+$/m, `contributor: anon-${hash}`);
-}
-
-/**
- * Extract the body (everything after the closing ---) from a skill file.
- */
-function extractBody(content) {
-  const match = content.match(/^---\s*\n[\s\S]*?\n---\s*\n([\s\S]*)$/);
-  return match ? match[1] : content;
-}
-
-function collectSkills(sessionIds) {
-  // Dedup by (name, domain, subdomain, memory_type) to preserve distinct skills
-  // while still collapsing true duplicates
-  const byKey = new Map();
-  for (const sid of sessionIds) {
-    const sessionDir = path.join(CACHE_DIR, sid);
-    if (!fs.existsSync(sessionDir)) continue;
-    for (const file of fs.readdirSync(sessionDir)) {
-      if (!file.endsWith('.md')) continue;
-      const src = path.join(sessionDir, file);
-      const content = fs.readFileSync(src, 'utf-8');
-      const name = extractField(content, 'name') || path.basename(file, '.md');
-      const domain = (extractField(content, 'domain') || '').toLowerCase();
-      const subdomain = (extractField(content, 'subdomain') || '').toLowerCase();
-      const memType = (extractField(content, 'memory_type') || '').toLowerCase();
-      const key = `${name.toLowerCase()}|${domain}|${subdomain}|${memType}`;
-      const bodyLen = extractBody(content).length;
-      const existing = byKey.get(key);
-      // Keep the longer body for duplicates (matches validate-skills.js collect behavior)
-      if (existing && existing.bodyLen >= bodyLen) continue;
-      byKey.set(key, { content, name, bodyLen });
-    }
-  }
-  // Convert to slug-keyed map, appending suffix on collision and
-  // disambiguating the frontmatter name so Codex can distinguish them
-  const skills = new Map();
-  for (const { content, name } of byKey.values()) {
-    const baseSlug = slugify(name);
-    let finalSlug = baseSlug;
-    let suffix = 2;
-    while (skills.has(finalSlug)) {
-      finalSlug = `${baseSlug}-${suffix++}`;
-    }
-    // If slug was disambiguated, also disambiguate the name in frontmatter
-    let finalContent = hashContributor(content);
-    let finalName = name;
-    if (finalSlug !== baseSlug) {
-      const memType = extractField(content, 'memory_type') || '';
-      finalName = memType ? `${name} (${memType})` : `${name} (${suffix - 1})`;
-      finalContent = finalContent.replace(
-        /^name:\s*.+$/m,
-        `name: "${finalName.replace(/"/g, '\\"')}"`
-      );
-    }
-    skills.set(finalSlug, { content: finalContent, name: finalName });
-  }
-  return skills;
-}
-
-function storeToTarget(targetName, skills) {
-  let installed = 0;
-  let skipped = 0;
-  let dir;
-
-  if (targetName === 'claude') {
-    // Claude Code: ~/.claude/commands/researchskills/<slug>.md
-    // Strip YAML frontmatter — Claude commands are plain markdown prompts
-    dir = path.join(os.homedir(), '.claude', 'commands', 'researchskills');
-    fs.mkdirSync(dir, { recursive: true });
-
-    for (const [slug, { content, name }] of skills) {
-      const body = extractBody(content).trim();
-      // Prepend a markdown title from the skill name
-      const commandContent = `# ${name}\n\n${body}\n`;
-      const dst = path.join(dir, `${slug}.md`);
-      if (fs.existsSync(dst) && fs.readFileSync(dst, 'utf-8') === commandContent) {
-        skipped++;
-        continue;
-      }
-      fs.writeFileSync(dst, commandContent);
-      installed++;
-    }
-  } else if (targetName === 'codex') {
-    // Codex: ~/.codex/skills/researchskills-<slug>/SKILL.md
-    dir = path.join(os.homedir(), '.codex', 'skills');
-    fs.mkdirSync(dir, { recursive: true });
-
-    for (const [slug, { content, name }] of skills) {
-      // Codex requires a `description` field in frontmatter to index/trigger skills.
-      // Build a retrievable description from domain + first body sentence.
-      let finalContent = content;
-      if (!extractField(content, 'description')) {
-        const memoryType = extractField(content, 'memory_type') || 'research';
-        const domain = extractField(content, 'domain') || '';
-        const subdomain = extractField(content, 'subdomain') || '';
-        const body = extractBody(content).trim();
-        // Extract first sentence (up to 120 chars) for trigger context
-        const firstLine = body.split(/\n/)[0] || '';
-        const snippet = firstLine.substring(0, 120).replace(/[#*_>]/g, '').trim();
-        const parts = [`${memoryType} skill`, domain, subdomain].filter(Boolean);
-        const rawDesc = snippet
-          ? `${parts.join(' / ')}: ${snippet}`
-          : `ResearchSkills ${parts.join(' / ')}: ${name}`;
-        // Escape quotes and backslashes for valid YAML
-        const desc = rawDesc.replace(/\\/g, '\\\\').replace(/"/g, '\\"');
-        finalContent = content.replace(/^(---\s*\n)/, `$1description: "${desc}"\n`);
-      }
-      const skillDir = path.join(dir, `researchskills-${slug}`);
-      fs.mkdirSync(skillDir, { recursive: true });
-      const dst = path.join(skillDir, 'SKILL.md');
-      if (fs.existsSync(dst) && fs.readFileSync(dst, 'utf-8') === finalContent) {
-        skipped++;
-        continue;
-      }
-      fs.writeFileSync(dst, finalContent);
-      installed++;
-    }
-  } else {
-    throw new Error(`Unknown target: "${targetName}". Must be "claude", "codex", or "both".`);
-  }
-
-  return { installed, skipped, dir };
-}
-
-// CLI
-if (require.main === module) {
-  const args = process.argv.slice(2);
-  let target = null;
-  let sessionIds = [];
-
-  for (let i = 0; i < args.length; i++) {
-    switch (args[i]) {
-      case '--target':     target = args[++i]; break;
-      case '--session-ids': sessionIds = args[++i].split(',').filter(Boolean); break;
-    }
-  }
-
-  if (!target || sessionIds.length === 0) {
-    console.error('Usage: store-local.js --target claude|codex|both --session-ids id1,id2,...');
-    process.exit(1);
-  }
-
-  const skills = collectSkills(sessionIds);
-
-  if (skills.size === 0) {
-    console.log('No cached skills found for the given session IDs.');
-    process.exit(0);
-  }
-
-  const targets = target === 'both' ? ['claude', 'codex'] : [target];
-  const results = {};
-
-  for (const t of targets) {
-    const result = storeToTarget(t, skills);
-    results[t] = result;
-    if (result.installed > 0) {
-      console.log(`✓ ${t}: ${result.installed} skill(s) installed to ${result.dir}`);
-    }
-    if (result.skipped > 0) {
-      console.log(`  ${t}: ${result.skipped} skill(s) already up-to-date`);
-    }
-  }
-
-  // Output JSON summary for the calling agent to parse
-  console.log(`\n${JSON.stringify({ total: skills.size, results })}`);
-}
-
-module.exports = { collectSkills, storeToTarget, slugify, extractField };
diff --git a/researchskills-extract/scripts/upload-skills.js b/researchskills-extract/scripts/upload-skills.js
deleted file mode 100755
index 93c601d..0000000
--- a/researchskills-extract/scripts/upload-skills.js
+++ /dev/null
@@ -1,297 +0,0 @@
-#!/usr/bin/env node
-/**
- * upload-skills.js
- *
- * Read skill .md files from a directory, parse frontmatter, and POST
- * each to the /api/skills endpoint on researchskills.ai.
- *
- * On failure, saves skills locally to ~/.researchskills/skills-fallback/
- * and exits non-zero so the caller can surface the error.
- *
- * Usage:
- *   upload-skills.js <skills_dir> [--no-open] [--headless] [--consent] [--api <url>]
- *
- * Exit codes:
- *   0 = all uploaded
- *   1 = usage error
- *   2 = upload failed
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const { execFileSync } = require('child_process');
-const { randomUUID } = require('crypto');
-
-const DEFAULT_API = 'https://researchskills.ai/api/skills';
-
-/**
- * Detect if running in a headless/SSH environment where a browser can't open.
- * Checks for SSH_CONNECTION, SSH_CLIENT, SSH_TTY (SSH session),
- * and missing DISPLAY on Linux (no X server).
- */
-function isHeadless() {
-  if (process.env.SSH_CONNECTION || process.env.SSH_CLIENT || process.env.SSH_TTY) {
-    return true;
-  }
-  // Linux without DISPLAY = no GUI
-  if (process.platform === 'linux' && !process.env.DISPLAY && !process.env.WAYLAND_DISPLAY) {
-    return true;
-  }
-  return false;
-}
-
-/**
- * Parse simple YAML frontmatter from markdown content.
- * Handles flat keys, one level of nesting, and arrays in [a, b] format.
- */
-function parseFrontmatter(content) {
-  const lines = content.split('\n');
-  if (lines[0].trim() !== '---') return null;
-
-  let endIdx = -1;
-  for (let i = 1; i < lines.length; i++) {
-    if (lines[i].trim() === '---') {
-      endIdx = i;
-      break;
-    }
-  }
-  if (endIdx === -1) return null;
-
-  const fm = {};
-  let currentKey = null;
-
-  for (let i = 1; i < endIdx; i++) {
-    const line = lines[i];
-    // Skip empty lines and comments
-    if (line.trim() === '' || line.trim().startsWith('#')) continue;
-
-    // Check for nested key (indented with spaces)
-    if (/^\s+\w/.test(line) && currentKey) {
-      const match = line.match(/^\s+(\w[\w_-]*):\s*(.*)/);
-      if (match) {
-        if (typeof fm[currentKey] !== 'object' || Array.isArray(fm[currentKey])) {
-          fm[currentKey] = {};
-        }
-        fm[currentKey][match[1]] = parseValue(match[2]);
-      }
-      continue;
-    }
-
-    // Top-level key
-    const match = line.match(/^(\w[\w_-]*):\s*(.*)/);
-    if (match) {
-      currentKey = match[1];
-      fm[currentKey] = parseValue(match[2]);
-    }
-  }
-
-  return { frontmatter: fm, body: lines.slice(endIdx + 1).join('\n') };
-}
-
-function parseValue(raw) {
-  const val = raw.trim();
-  if (val === '') return '';
-  // Array in [a, b] format
-  if (val.startsWith('[') && val.endsWith(']')) {
-    return val
-      .slice(1, -1)
-      .split(',')
-      .map((s) => s.trim())
-      .filter((s) => s.length > 0);
-  }
-  // Strip quotes
-  if ((val.startsWith('"') && val.endsWith('"')) ||
-      (val.startsWith("'") && val.endsWith("'"))) {
-    return val.slice(1, -1);
-  }
-  return val;
-}
-
-/**
- * POST JSON to an API endpoint using Node 18+ built-in fetch().
- */
-async function postSkill(apiUrl, payload) {
-  const body = JSON.stringify(payload);
-  const res = await fetch(apiUrl, {
-    method: 'POST',
-    headers: { 'Content-Type': 'application/json' },
-    body,
-    signal: AbortSignal.timeout(30000),
-  });
-
-  if (!res.ok) {
-    const text = await res.text().catch(() => '(empty body)');
-    throw new Error(`HTTP ${res.status}: ${text.substring(0, 300)}`);
-  }
-
-  return res.json();
-}
-
-function fallbackSave(skills) {
-  const dir = path.join(os.homedir(), '.researchskills', 'skills-fallback');
-  fs.mkdirSync(dir, { recursive: true });
-
-  for (const skill of skills) {
-    const safeName = String(skill.name || 'unknown')
-      .replace(/[^a-zA-Z0-9-_]/g, '_')
-      .substring(0, 80);
-    const fallbackPath = path.join(dir, `${safeName}.json`);
-    fs.writeFileSync(fallbackPath, JSON.stringify(skill, null, 2) + '\n');
-  }
-
-  return dir;
-}
-
-function openBrowser(url) {
-  try {
-    if (process.platform === 'darwin') {
-      execFileSync('open', [url], { stdio: 'ignore' });
-    } else if (process.platform === 'win32') {
-      execFileSync('cmd', ['/c', 'start', '', url], { stdio: 'ignore' });
-    } else {
-      execFileSync('xdg-open', [url], { stdio: 'ignore' });
-    }
-    return true;
-  } catch (err) {
-    return false;
-  }
-}
-
-async function uploadSkills(skillsDir, options = {}) {
-  const apiUrl = options.apiUrl || DEFAULT_API;
-  const batchId = randomUUID();
-  const headless = options.headless || false;
-  const consent = options.consent || false;
-
-  const files = fs.readdirSync(skillsDir).filter((f) => f.endsWith('.md'));
-  if (files.length === 0) {
-    return { ok: false, error: new Error('No .md files found in directory'), skills: [] };
-  }
-
-  const results = [];
-  const failedSkills = [];
-  let allOk = true;
-
-  if (headless) {
-    console.log(`[headless] Batch ${batchId} — uploading ${files.length} skill(s)${consent ? ' with consent' : ''}`);
-  } else {
-    console.log(`Batch ${batchId} — uploading ${files.length} skill(s)`);
-  }
-
-  for (const file of files) {
-    const filePath = path.join(skillsDir, file);
-    const content = fs.readFileSync(filePath, 'utf-8');
-    const parsed = parseFrontmatter(content);
-
-    if (!parsed) {
-      console.error(`\u2717 ${file}: missing or malformed frontmatter`);
-      allOk = false;
-      continue;
-    }
-
-    const payload = { ...parsed.frontmatter, body: parsed.body, batch_id: batchId };
-    if (options.projectSlug) payload.project_slug = options.projectSlug;
-    if (options.projectName) payload.project_name = options.projectName;
-    if (consent) payload.consent = true;
-
-    try {
-      const response = await postSkill(apiUrl, payload);
-      console.log(`\u2713 ${file}`);
-      results.push({ file, ok: true, response });
-    } catch (err) {
-      console.error(`\u2717 ${file}: ${err.message}`);
-      allOk = false;
-      failedSkills.push(payload);
-      results.push({ file, ok: false, error: err.message });
-    }
-  }
-
-  return { ok: allOk, results, failedSkills, batchId };
-}
-
-// CLI
-if (require.main === module) {
-  const args = process.argv.slice(2);
-  if (args.length < 1 || args[0] === '--help' || args[0] === '-h') {
-    console.error('Usage: upload-skills.js <skills_dir> [--no-open] [--headless] [--consent] [--api <url>]');
-    process.exit(args.length < 1 ? 1 : 0);
-  }
-
-  const skillsDir = path.resolve(args[0]);
-  const headless = args.includes('--headless') || isHeadless();
-  const consent = args.includes('--consent');
-  const noOpen = args.includes('--no-open') || headless;
-  const apiIdx = args.indexOf('--api');
-  const apiUrl = apiIdx !== -1 && args[apiIdx + 1] ? args[apiIdx + 1] : DEFAULT_API;
-  const projectSlugIdx = args.indexOf('--project-slug');
-  const projectSlug = projectSlugIdx !== -1 && args[projectSlugIdx + 1] ? args[projectSlugIdx + 1] : null;
-  const projectNameIdx = args.indexOf('--project-name');
-  const projectName = projectNameIdx !== -1 && args[projectNameIdx + 1] ? args[projectNameIdx + 1] : null;
-
-  if (headless) {
-    console.log('[headless] SSH/remote environment detected — browser open disabled');
-  }
-
-  if (!fs.existsSync(skillsDir) || !fs.statSync(skillsDir).isDirectory()) {
-    console.error(`Error: directory not found: ${skillsDir}`);
-    process.exit(1);
-  }
-
-  uploadSkills(skillsDir, { apiUrl, projectSlug, projectName, headless, consent }).then((result) => {
-    if (result.ok) {
-      const uploaded = result.results.filter((r) => r.ok);
-      const ids = uploaded
-        .map((r) => r.response && (r.response.id || r.response.skill_id))
-        .filter(Boolean);
-
-      const baseUrl = apiUrl.replace(/\/api\/skills$/, '');
-      const batchReviewUrl = `${baseUrl}/review/batch/${result.batchId}`;
-
-      const summary = {
-        count: uploaded.length,
-        batchId: result.batchId,
-        batchReviewUrl,
-      };
-      if (ids.length > 0) summary.ids = ids;
-
-      console.log(`RESULT=${JSON.stringify(summary)}`);
-
-      if (headless) {
-        console.log(`\u2713 ${uploaded.length} skill(s) uploaded${consent ? ' with consent' : ''}`);
-        console.log(`  Review your skills (from any browser): ${batchReviewUrl}`);
-        console.log('  Sign in with GitHub on the review page to claim credit and submit.');
-      } else if (!noOpen) {
-        if (openBrowser(batchReviewUrl)) {
-          console.log('\u2713 Opened batch review page in browser');
-        } else {
-          console.log(`  (Could not open browser \u2014 visit manually: ${batchReviewUrl})`);
-        }
-      }
-
-      process.exit(0);
-    } else {
-      if (result.failedSkills && result.failedSkills.length > 0) {
-        const fallbackDir = fallbackSave(result.failedSkills);
-        console.error(`\u26A0 Upload failed for ${result.failedSkills.length} skill(s)`);
-        console.error(`  Saved locally to: ${fallbackDir}`);
-        console.error('  You can upload manually later.');
-      } else if (result.error) {
-        console.error(`\u26A0 ${result.error.message}`);
-      }
-      process.exit(2);
-    }
-  });
-}
-
-module.exports = {
-  uploadSkills,
-  postSkill,
-  parseFrontmatter,
-  fallbackSave,
-  openBrowser,
-  isHeadless,
-  DEFAULT_API,
-};
diff --git a/researchskills-extract/scripts/validate-skills.js b/researchskills-extract/scripts/validate-skills.js
deleted file mode 100644
index b97ea1e..0000000
--- a/researchskills-extract/scripts/validate-skills.js
+++ /dev/null
@@ -1,466 +0,0 @@
-#!/usr/bin/env node
-/**
- * validate-skills.js
- *
- * Helper for the /researchskills-extract AI phase. The AI reads a formatted
- * session and extracts cognitive memory-type skills as markdown files.
- * This script validates the skill markdown (frontmatter + required
- * sections) and caches validated skills per session.
- *
- * Subcommands:
- *   save <session_id> <file1.md> [file2.md ...]
- *       Validate each skill file and copy to cache dir.
- *
- *   list [session_id]
- *       Print JSON array of cached sessions, or skills within a session.
- *
- *   collect <output_dir> [session_ids_csv]
- *       Copy cached skills to output dir. Print count.
- *
- *   is-cached <session_id>
- *       Exit 0 if any cached skills exist for this session, 1 otherwise.
- *
- * Cache location: ~/.researchskills/cache/skills/<session_id>/
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-const crypto = require('crypto');
-
-const CACHE_DIR = path.join(os.homedir(), '.researchskills', 'cache', 'skills');
-
-const VALID_MEMORY_TYPES = new Set(['procedural', 'semantic', 'episodic']);
-
-const VALID_SUBTYPES = {
-  procedural: new Set(['tie', 'no-change', 'constraint-failure', 'operator-fail']),
-  semantic: new Set(['frontier', 'non-public', 'correction']),
-  episodic: new Set(['failure', 'adaptation', 'anomalous']),
-};
-
-const REQUIRED_SECTIONS = {
-  procedural: ['When', 'Decision', 'Local Verifiers', 'Failure Handling'],
-  semantic: ['Fact', 'Evidence'],
-  episodic: ['Situation', 'Action', 'Outcome', 'Retrieval Cues'],
-};
-
-const REQUIRED_FRONTMATTER_FIELDS = [
-  'name', 'memory_type', 'subtype', 'domain', 'subdomain', 'contributor',
-];
-
-// Engineering tags — if a skill's tags hit 2+ of these, reject as engineering
-const ENGINEERING_TAGS = new Set([
-  'canvas', 'canvas-2d', 'css', 'frontend', 'deployment', 'railway', 'vercel',
-  'docker', 'supabase', 'database', 'npm', 'webpack', 'vite', 'browser',
-  'rendering', 'ui', 'devops', 'ci-cd', 'hosting', 'dns', 'tls',
-  'postgresql', 'redis', 'connection-pooling', 'responsive-design',
-  'svg', 'animations', 'websocket', 'localstorage', 'typescript-config',
-  'github-actions', 'codeowners', 'pr-workflows', 'git',
-  'ui-performance', 'fractal-geometry', 'browser-quirks', 'color-formats',
-  'environment-variables', 'nodejs',
-  'mermaid', 'diagram', 'visualization', 'd3', 'chart',
-  'oauth', 'jwt', 'authentication', 'session-management',
-  'logging', 'debugging', 'error-handling', 'stack-trace',
-  'naming-convention', 'terminology', 'ux-copy', 'localization',
-]);
-
-const MIN_LLM_SCORE = 3;
-
-/**
- * Parse simple YAML frontmatter from markdown content.
- * Handles flat keys, one level of nesting, and arrays in [a, b] format.
- */
-function parseFrontmatter(content) {
-  const lines = content.split('\n');
-  if (lines[0].trim() !== '---') return null;
-
-  let endIdx = -1;
-  for (let i = 1; i < lines.length; i++) {
-    if (lines[i].trim() === '---') {
-      endIdx = i;
-      break;
-    }
-  }
-  if (endIdx === -1) return null;
-
-  const fm = {};
-  let currentKey = null;
-
-  for (let i = 1; i < endIdx; i++) {
-    const line = lines[i];
-    // Skip empty lines and comments
-    if (line.trim() === '' || line.trim().startsWith('#')) continue;
-
-    // Check for nested key (indented with spaces)
-    if (/^\s+\w/.test(line) && currentKey) {
-      const match = line.match(/^\s+(\w[\w_-]*):\s*(.*)/);
-      if (match) {
-        if (typeof fm[currentKey] !== 'object' || Array.isArray(fm[currentKey])) {
-          fm[currentKey] = {};
-        }
-        fm[currentKey][match[1]] = parseValue(match[2]);
-      }
-      continue;
-    }
-
-    // Top-level key
-    const match = line.match(/^(\w[\w_-]*):\s*(.*)/);
-    if (match) {
-      currentKey = match[1];
-      fm[currentKey] = parseValue(match[2]);
-    }
-  }
-
-  return { frontmatter: fm, body: lines.slice(endIdx + 1).join('\n') };
-}
-
-function parseValue(raw) {
-  const val = raw.trim();
-  if (val === '') return '';
-  // Array in [a, b] format
-  if (val.startsWith('[') && val.endsWith(']')) {
-    return val
-      .slice(1, -1)
-      .split(',')
-      .map((s) => s.trim())
-      .filter((s) => s.length > 0);
-  }
-  // Strip quotes
-  if ((val.startsWith('"') && val.endsWith('"')) ||
-      (val.startsWith("'") && val.endsWith("'"))) {
-    return val.slice(1, -1);
-  }
-  return val;
-}
-
-/**
- * Validate a skill markdown file. Returns an array of error strings.
- */
-function validateSkill(content, filename) {
-  const errors = [];
-  const parsed = parseFrontmatter(content);
-
-  if (!parsed) {
-    errors.push(`${filename}: missing or malformed YAML frontmatter`);
-    return errors;
-  }
-
-  const { frontmatter: fm, body } = parsed;
-
-  // Check required frontmatter fields
-  for (const field of REQUIRED_FRONTMATTER_FIELDS) {
-    if (!fm[field] || (typeof fm[field] === 'string' && fm[field].trim() === '')) {
-      errors.push(`${filename}: missing required frontmatter field "${field}"`);
-    }
-  }
-
-  // Validate memory_type
-  const memType = fm.memory_type;
-  if (memType && !VALID_MEMORY_TYPES.has(memType)) {
-    errors.push(`${filename}: invalid memory_type "${memType}"`);
-    return errors; // Can't validate further without valid type
-  }
-
-  // Validate subtype against memory_type
-  if (memType && fm.subtype) {
-    const validSubs = VALID_SUBTYPES[memType];
-    if (validSubs && !validSubs.has(fm.subtype)) {
-      errors.push(
-        `${filename}: invalid subtype "${fm.subtype}" for memory_type "${memType}" ` +
-        `(valid: ${[...validSubs].join(', ')})`
-      );
-    }
-  }
-
-  // Check required body sections for the memory type
-  if (memType && REQUIRED_SECTIONS[memType]) {
-    const requiredSections = REQUIRED_SECTIONS[memType];
-    for (const section of requiredSections) {
-      const pattern = new RegExp(`^##\\s+${escapeRegExp(section)}\\s*$`, 'm');
-      if (!pattern.test(body)) {
-        errors.push(`${filename}: missing required section "## ${section}"`);
-      }
-    }
-  }
-
-  // Reject low llm_score
-  const score = Number(fm.llm_score);
-  if (!isNaN(score) && score < MIN_LLM_SCORE) {
-    errors.push(`${filename}: rejected (llm_score ${score} < ${MIN_LLM_SCORE})`);
-  }
-
-  // Reject engineering content by tag overlap
-  const tags = Array.isArray(fm.tags) ? fm.tags.map(t => String(t).toLowerCase()) : [];
-  const engHits = tags.filter(t => ENGINEERING_TAGS.has(t));
-  if (engHits.length >= 2) {
-    errors.push(`${filename}: rejected as engineering (tags: ${engHits.join(', ')})`);
-  }
-
-  // PII detection — reject skills with leaked personal/private data
-  const bodyText = body || '';
-  const SAFE_URL_HOSTS = /^https?:\/\/(?:arxiv\.org|doi\.org|github\.com|en\.wikipedia\.org|researchskills\.ai)/;
-  const PLACEHOLDER_EMAIL = /^[a-z]@[a-z]\.[a-z]+$/; // e.g. x@y.com
-  const piiChecks = [
-    {
-      pattern: /https?:\/\/[^\s)"`]+/g,
-      label: 'private URL',
-      filter: (m) => !SAFE_URL_HOSTS.test(m),
-    },
-    { pattern: /[\u4e00-\u9fff]{4,}/g, label: 'CJK text (possible direct quote)', filter: null },
-    {
-      pattern: /[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}/g,
-      label: 'email address',
-      filter: (m) => !PLACEHOLDER_EMAIL.test(m.toLowerCase()),
-    },
-  ];
-  for (const { pattern, label, filter } of piiChecks) {
-    const matches = bodyText.match(pattern) || [];
-    const real = filter ? matches.filter(filter) : matches;
-    if (real.length > 0) {
-      errors.push(`${filename}: potential PII — ${label}: "${real[0].substring(0, 60)}"`);
-    }
-  }
-
-  return errors;
-}
-
-function escapeRegExp(str) {
-  return str.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-}
-
-function hashContributor(content) {
-  const match = content.match(/^contributor:\s*(.+)$/m);
-  if (!match) return content;
-  const raw = match[1].trim().replace(/^["']|["']$/g, '');
-  const hash = crypto.createHash('sha256').update(raw).digest('hex').substring(0, 8);
-  return content.replace(/^contributor:\s*.+$/m, `contributor: anon-${hash}`);
-}
-
-function sessionCacheDir(sessionId) {
-  return path.join(CACHE_DIR, sessionId);
-}
-
-function saveSkill(sessionId, filePath) {
-  let content = fs.readFileSync(filePath, 'utf-8');
-  const filename = path.basename(filePath);
-  const errors = validateSkill(content, filename);
-
-  if (errors.length > 0) {
-    return { ok: false, errors, filename };
-  }
-
-  const dest = sessionCacheDir(sessionId);
-  fs.mkdirSync(dest, { recursive: true });
-  fs.writeFileSync(path.join(dest, filename), content);
-  return { ok: true, errors: [], filename };
-}
-
-function isCached(sessionId) {
-  const dir = sessionCacheDir(sessionId);
-  if (!fs.existsSync(dir)) return false;
-  const entries = fs.readdirSync(dir);
-  // Cached if skills exist OR .done marker present (segment produced 0 skills)
-  return entries.some((f) => f.endsWith('.md') || f.startsWith('.done'));
-}
-
-function listCachedSessions() {
-  if (!fs.existsSync(CACHE_DIR)) return [];
-  return fs
-    .readdirSync(CACHE_DIR)
-    .filter((f) => {
-      const full = path.join(CACHE_DIR, f);
-      return fs.statSync(full).isDirectory();
-    });
-}
-
-function listSkillsInSession(sessionId) {
-  const dir = sessionCacheDir(sessionId);
-  if (!fs.existsSync(dir)) return [];
-  return fs.readdirSync(dir).filter((f) => f.endsWith('.md'));
-}
-
-function collectSkills(outputDir, filterIds) {
-  const sessions = listCachedSessions();
-  const filter = filterIds && filterIds.length > 0 ? new Set(filterIds) : null;
-  let count = 0;
-  let duplicates = 0;
-
-  fs.mkdirSync(outputDir, { recursive: true });
-
-  // Deduplicate by skill name across segments/sessions.
-  // For same-name skills, keep the one with longer body content.
-  const seenNames = new Map(); // name → { file, srcPath, bodyLen }
-
-  for (const sid of sessions) {
-    if (filter && !filter.has(sid)) continue;
-    const srcDir = sessionCacheDir(sid);
-    const files = fs.readdirSync(srcDir).filter((f) => f.endsWith('.md'));
-    for (const file of files) {
-      const srcPath = path.join(srcDir, file);
-      const content = fs.readFileSync(srcPath, 'utf-8');
-      const parsed = parseFrontmatter(content);
-      const name = parsed && parsed.frontmatter && parsed.frontmatter.name
-        ? String(parsed.frontmatter.name).trim().toLowerCase()
-        : null;
-
-      if (name && seenNames.has(name)) {
-        // Duplicate — keep the longer one
-        const existing = seenNames.get(name);
-        const bodyLen = parsed.body ? parsed.body.length : 0;
-        if (bodyLen > existing.bodyLen) {
-          seenNames.set(name, { file, srcPath, bodyLen });
-        }
-        duplicates++;
-        continue;
-      }
-
-      const bodyLen = parsed && parsed.body ? parsed.body.length : 0;
-      if (name) {
-        seenNames.set(name, { file, srcPath, bodyLen });
-      } else {
-        // No parseable name — copy directly (don't deduplicate)
-        fs.copyFileSync(srcPath, path.join(outputDir, file));
-        count++;
-      }
-    }
-  }
-
-  // Copy deduplicated skills, hashing contributor for de-identification
-  for (const { file, srcPath } of seenNames.values()) {
-    let content = fs.readFileSync(srcPath, 'utf-8');
-    content = hashContributor(content);
-    fs.writeFileSync(path.join(outputDir, file), content);
-    count++;
-  }
-
-  if (duplicates > 0) {
-    console.log(`  Deduplicated: ${duplicates} duplicate skills removed`);
-  }
-  return count;
-}
-
-function revalidateCache(filterSessionId) {
-  const sessions = filterSessionId ? [filterSessionId] : listCachedSessions();
-  let total = 0, removed = 0;
-
-  for (const sid of sessions) {
-    const dir = sessionCacheDir(sid);
-    if (!fs.existsSync(dir)) continue;
-    const files = fs.readdirSync(dir).filter((f) => f.endsWith('.md'));
-
-    for (const file of files) {
-      total++;
-      const filePath = path.join(dir, file);
-      const content = fs.readFileSync(filePath, 'utf-8');
-      const errors = validateSkill(content, file);
-      if (errors.length > 0) {
-        removed++;
-        fs.unlinkSync(filePath);
-        console.log(`  ✗ ${sid}/${file}`);
-        errors.forEach((e) => console.log(`    ${e}`));
-      }
-    }
-
-    // Clean up empty session dirs
-    if (fs.existsSync(dir) && fs.readdirSync(dir).filter(f => f.endsWith('.md')).length === 0) {
-      fs.rmdirSync(dir);
-    }
-  }
-
-  console.log(`\nRevalidated ${total} skills: ${removed} removed, ${total - removed} kept.`);
-  return { total, removed };
-}
-
-function printUsage() {
-  console.error('Usage: validate-skills.js <command> [args...]');
-  console.error('');
-  console.error('Commands:');
-  console.error('  save <session_id> <file1.md> [file2.md ...]');
-  console.error('      Validate skill files and cache them.');
-  console.error('  list [session_id]');
-  console.error('      List cached sessions or skills within a session.');
-  console.error('  collect <output_dir> [session_ids_csv]');
-  console.error('      Copy cached skills to output dir.');
-  console.error('  is-cached <session_id>');
-  console.error('      Exit 0 if cached skills exist, 1 otherwise.');
-  console.error('  revalidate [session_id]');
-  console.error('      Re-validate cached skills with current rules. Removes failures.');
-}
-
-// CLI
-if (require.main === module) {
-  const args = process.argv.slice(2);
-  const cmd = args[0];
-
-  try {
-    if (cmd === 'save') {
-      if (args.length < 3) {
-        printUsage();
-        process.exit(1);
-      }
-      const sessionId = args[1];
-      const files = args.slice(2).map((f) => path.resolve(f));
-      let allOk = true;
-
-      for (const filePath of files) {
-        const result = saveSkill(sessionId, filePath);
-        if (result.ok) {
-          console.log(`✓ ${result.filename}`);
-        } else {
-          allOk = false;
-          console.error(`✗ ${result.filename}`);
-          result.errors.forEach((e) => console.error(`  - ${e}`));
-        }
-      }
-
-      if (!allOk) process.exit(1);
-    } else if (cmd === 'list') {
-      if (args[1]) {
-        console.log(JSON.stringify(listSkillsInSession(args[1]), null, 2));
-      } else {
-        console.log(JSON.stringify(listCachedSessions(), null, 2));
-      }
-    } else if (cmd === 'collect') {
-      if (args.length < 2) {
-        printUsage();
-        process.exit(1);
-      }
-      const outputDir = path.resolve(args[1]);
-      const filter = args[2] ? args[2].split(',').map((s) => s.trim()) : null;
-      const count = collectSkills(outputDir, filter);
-      console.log(
-        `✓ Collected ${count} skills → ${path.relative(process.cwd(), outputDir)}`
-      );
-    } else if (cmd === 'is-cached') {
-      if (args.length < 2) {
-        printUsage();
-        process.exit(1);
-      }
-      process.exit(isCached(args[1]) ? 0 : 1);
-    } else if (cmd === 'revalidate') {
-      revalidateCache(args[1] || null);
-    } else {
-      printUsage();
-      process.exit(1);
-    }
-  } catch (err) {
-    console.error(`Error: ${err.message}`);
-    process.exit(1);
-  }
-}
-
-module.exports = {
-  parseFrontmatter,
-  validateSkill,
-  saveSkill,
-  isCached,
-  listCachedSessions,
-  listSkillsInSession,
-  collectSkills,
-  CACHE_DIR,
-  VALID_MEMORY_TYPES,
-  VALID_SUBTYPES,
-  REQUIRED_SECTIONS,
-};
diff --git a/researchskills-extract/templates/report.html b/researchskills-extract/templates/report.html
deleted file mode 100644
index fc74055..0000000
--- a/researchskills-extract/templates/report.html
+++ /dev/null
@@ -1,325 +0,0 @@
-<!DOCTYPE html>
-<html lang="en">
-<head>
-<meta charset="UTF-8">
-<meta name="viewport" content="width=device-width, initial-scale=1.0">
-<title>ResearchSkills — Know-How Extraction Report</title>
-<style>
-*{margin:0;padding:0;box-sizing:border-box}
-body{font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',sans-serif;background:#0d1117;color:#e6edf3;line-height:1.6}
-.container{max-width:960px;margin:0 auto;padding:2rem}
-.header{text-align:center;padding:3rem 0 2rem;border-bottom:1px solid #30363d;margin-bottom:2rem}
-.header h1{font-size:2rem;margin-bottom:.5rem}
-.header .subtitle{color:#8b949e;font-size:1rem}
-.stats{display:flex;justify-content:center;gap:2rem;margin-top:1rem}
-.stat{text-align:center}
-.stat-value{font-size:1.5rem;font-weight:bold;color:#58a6ff}
-.stat-label{font-size:.8rem;color:#8b949e}
-.author-bar{display:flex;justify-content:center;gap:1.5rem;margin-bottom:2rem;flex-wrap:wrap}
-.author-field{display:flex;align-items:center;gap:.5rem;color:#8b949e;font-size:.9rem}
-.author-field input{background:#161b22;border:1px solid #30363d;color:#e6edf3;padding:6px 12px;border-radius:4px;font-size:.9rem;width:200px}
-.project{background:#161b22;border:1px solid #30363d;border-radius:8px;margin-bottom:1.5rem;overflow:hidden}
-.project-header{padding:1rem 1.5rem;background:#1c2129;border-bottom:1px solid #30363d;display:flex;justify-content:space-between;align-items:center;flex-wrap:wrap;gap:.5rem}
-.project-title{font-size:1.2rem;font-weight:600}
-.project-domain{background:#388bfd26;color:#58a6ff;padding:2px 8px;border-radius:12px;font-size:.8rem}
-.skill{border-bottom:1px solid #21262d;padding:1rem 1.5rem}
-.skill:last-child{border-bottom:none}
-.skill.rejected{opacity:.35}
-.skill-top{display:flex;align-items:center;gap:.75rem}
-.skill-cb{width:18px;height:18px;accent-color:#238636;cursor:pointer;flex-shrink:0}
-.badge{font-size:.7rem;padding:2px 8px;border-radius:12px;font-weight:500;white-space:nowrap}
-.c01{background:#1f6feb26;color:#58a6ff}.c02{background:#8957e526;color:#bc8cff}
-.c03{background:#f7883026;color:#f0883e}.c04{background:#23863626;color:#3fb950}
-.c05{background:#da363626;color:#f85149}.c06{background:#1f6feb26;color:#79c0ff}
-.c07{background:#8957e526;color:#d2a8ff}.c08{background:#f7883026;color:#ffa657}
-.c09{background:#23863626;color:#56d364}.c10{background:#da363626;color:#ff7b72}
-.skill-title{font-weight:600;flex:1;min-width:0}
-.skill-conf{font-size:.75rem;color:#8b949e;white-space:nowrap}
-.skill-desc{color:#8b949e;font-size:.9rem;margin:.5rem 0 0 2.25rem}
-.expand-btn{font-size:.8rem;color:#58a6ff;cursor:pointer;margin-left:2.25rem;margin-top:.25rem;background:none;border:none;padding:0}
-.expand-btn:hover{text-decoration:underline}
-.gh-btn{font-size:.75rem;background:#238636;color:#fff;border:none;padding:3px 10px;border-radius:4px;cursor:pointer;margin-left:.5rem;text-decoration:none;display:inline-block}
-.gh-btn:hover{background:#2ea043}
-.skill-actions{margin:.5rem 0 0 2.25rem;display:flex;gap:.5rem;align-items:center}
-.detail{display:none;margin:1rem 0 0 2.25rem;padding:1rem;background:#0d1117;border-radius:6px;border:1px solid #30363d;font-size:.85rem}
-.detail.open{display:block}
-.detail h4{color:#58a6ff;margin:.75rem 0 .25rem;font-size:.85rem}
-.detail h4:first-child{margin-top:0}
-.editable{border:1px solid transparent;border-radius:3px;padding:2px 4px;transition:border-color .2s;outline:none;min-height:1.5em}
-.editable:hover{border-color:#30363d}
-.editable:focus{border-color:#58a6ff;background:#161b22}
-.detail ul{padding-left:1.25rem;margin:.25rem 0}
-.detail li{margin:.25rem 0}
-.editable strong{color:#e6edf3;font-weight:600}
-.editable code{background:#30363d;padding:1px 5px;border-radius:3px;font-size:.85em}
-.submit-section{margin-top:2rem;padding:2rem;background:#161b22;border:1px solid #30363d;border-radius:8px;text-align:center}
-.submit-btn{background:#238636;color:#fff;border:none;padding:12px 32px;border-radius:6px;font-size:1rem;font-weight:600;cursor:pointer;transition:background .2s}
-.submit-btn:hover:not(:disabled){background:#2ea043}
-.submit-btn:disabled{opacity:.4;cursor:not-allowed}
-.submit-info{color:#8b949e;margin-bottom:1rem}
-.consent-banner{background:#1c1f26;border:1px solid #f0883e44;border-radius:8px;padding:1.25rem 1.5rem;margin-bottom:1.25rem;text-align:left}
-.consent-banner p{color:#e6edf3;font-size:.9rem;margin-bottom:.75rem;line-height:1.5}
-.consent-banner label{display:flex;align-items:flex-start;gap:.5rem;color:#e6edf3;font-size:.85rem;cursor:pointer;line-height:1.5}
-.consent-banner input[type="checkbox"]{margin-top:3px;accent-color:#238636;width:16px;height:16px;flex-shrink:0}
-.consent-banner .consent-details{color:#8b949e;font-size:.8rem;margin-top:.5rem;padding-left:1.5rem}
-.gh-btn:disabled{opacity:.4;cursor:not-allowed}
-.output{margin-top:1rem;text-align:left;display:none}
-.output pre{background:#0d1117;padding:1rem;border-radius:6px;overflow-x:auto;font-size:.8rem;white-space:pre-wrap;border:1px solid #30363d;max-height:400px;overflow-y:auto}
-.copy-btn{background:#30363d;color:#e6edf3;border:1px solid #484f58;padding:6px 16px;border-radius:6px;cursor:pointer;margin-top:.5rem;font-size:.85rem}
-.copy-btn:hover{background:#484f58}
-.footer{text-align:center;padding:2rem 0;color:#484f58;font-size:.85rem}
-.footer a{color:#58a6ff;text-decoration:none}
-@media(max-width:640px){.container{padding:1rem}.stats{gap:1rem}.project-header{flex-direction:column;align-items:flex-start}}
-</style>
-</head>
-<body>
-<div class="container">
-  <div class="header">
-    <h1>🏛️ ResearchSkills</h1>
-    <div class="subtitle">Know-How Extraction Report</div>
-    <div class="stats">
-      <div class="stat"><div class="stat-value" id="s-proj">0</div><div class="stat-label">Projects</div></div>
-      <div class="stat"><div class="stat-value" id="s-skill">0</div><div class="stat-label">Skills Found</div></div>
-      <div class="stat"><div class="stat-value" id="s-sess">0</div><div class="stat-label">Sessions Analyzed</div></div>
-    </div>
-  </div>
-  <div class="author-bar">
-    <div class="author-field"><label for="author-name">Name:</label> <input id="author-name" type="text" placeholder="e.g. Albert Einstein"></div>
-    <div class="author-field"><label for="author-inst">Institution:</label> <input id="author-inst" type="text" placeholder="e.g. ETH Zürich Physics"></div>
-    <div class="author-field"><label for="author-role">Role:</label> <select id="author-role" style="background:#161b22;border:1px solid #30363d;color:#e6edf3;padding:6px 12px;border-radius:4px;font-size:.9rem">
-      <option value="">Select...</option>
-      <option value="Undergraduate Student">Undergraduate Student</option>
-      <option value="Master's Student">Master's Student</option>
-      <option value="PhD Candidate">PhD Candidate</option>
-      <option value="Postdoctoral Researcher">Postdoctoral Researcher</option>
-      <option value="Research Scientist">Research Scientist</option>
-      <option value="Research Engineer">Research Engineer</option>
-      <option value="Assistant Professor">Assistant Professor</option>
-      <option value="Associate Professor">Associate Professor</option>
-      <option value="Professor">Professor</option>
-      <option value="Principal Investigator (PI)">Principal Investigator (PI)</option>
-      <option value="Industry Researcher">Industry Researcher</option>
-      <option value="Independent Researcher">Independent Researcher</option>
-    </select></div>
-  </div>
-  <div id="projects"></div>
-  <div class="submit-section">
-    <div class="consent-banner">
-      <p>Nothing is uploaded until you explicitly agree. Review and edit your skills above first.</p>
-      <label>
-        <input type="checkbox" id="consent-cb" onchange="onConsentChange()">
-        I agree to submit my selected skills to ResearchSkills for reviewer review. Submitted skills will be stored on the ResearchSkills server and reviewed by a maintainer before publication. Skills are licensed under CC-BY 4.0.
-      </label>
-      <div class="consent-details">Your name, institution, and role will be attached as contributor metadata. All skills are de-identified during extraction, but please verify no private information remains before submitting.</div>
-    </div>
-    <div class="submit-info">After agreeing above, click "Submit" on individual skills or submit all accepted skills at once.</div>
-    <button class="submit-btn" id="submit-all-btn" onclick="submitAll()" disabled>Submit All Accepted to GitHub</button>
-    <div class="output" id="output">
-      <p style="color:#3fb950;margin-bottom:.5rem" id="submit-msg"></p>
-    </div>
-  </div>
-  <div class="footer">Generated by <a href="https://github.com/ScienceIntelligence/ResearchSkills">/researchskills-extract</a> · <span id="rdate"></span></div>
-</div>
-
-<script>
-// __REPORT_DATA__ is replaced by the prompt with actual extracted JSON
-const DATA = __REPORT_DATA__;
-
-let consentGiven = false;
-function onConsentChange(){
-  consentGiven = document.getElementById('consent-cb').checked;
-  document.getElementById('submit-all-btn').disabled = !consentGiven;
-  document.querySelectorAll('.gh-btn').forEach(btn => btn.disabled = !consentGiven);
-}
-
-const CAT = {
-  '01-literature-search':'Literature Search','02-hypothesis-and-ideation':'Hypothesis & Ideation',
-  '03-math-and-modeling':'Math & Modeling','04-experiment-planning':'Experiment Planning',
-  '05-data-acquisition':'Data Acquisition','06-coding-and-execution':'Coding & Execution',
-  '07-result-analysis':'Result Analysis','08-reusable-tooling':'Reusable Tooling',
-  '09-paper-writing':'Paper Writing','10-review-and-rebuttal':'Review & Rebuttal'
-};
-
-const projects = DATA.projects || [];
-document.getElementById('author-name').value = DATA.author || '';
-document.getElementById('author-inst').value = DATA.institution || '';
-if(DATA.role) document.getElementById('author-role').value = DATA.role;
-document.getElementById('rdate').textContent = DATA.date || new Date().toISOString().slice(0,10);
-document.getElementById('s-proj').textContent = projects.length;
-document.getElementById('s-skill').textContent = projects.reduce((a,p) => a + p.skills.length, 0);
-document.getElementById('s-sess').textContent = DATA.total_sessions || 0;
-
-const root = document.getElementById('projects');
-
-function esc(s){const d=document.createElement('div');d.textContent=s||'';return d.innerHTML}
-function md(s){
-  if(!s) return '';
-  return esc(s)
-    .replace(/\*\*(.+?)\*\*/g,'<strong>$1</strong>')
-    .replace(/`(.+?)`/g,'<code style="background:#30363d;padding:1px 5px;border-radius:3px;font-size:.85em">$1</code>')
-    .replace(/\n- (.+)/g,'\n<li>$1</li>')
-    .replace(/(<li>.*<\/li>)/gs,'<ul>$1</ul>')
-    .replace(/\n/g,'<br>');
-}
-function mkEditable(text, cls){return `<span class="editable ${cls||''}" contenteditable="true">${md(text)}</span>`}
-function mkEditableList(arr, cls){
-  if(!arr||!arr.length) return '<ul><li><span class="editable" contenteditable="true">N/A</span></li></ul>';
-  return '<ul>'+arr.map(item=>`<li><span class="editable ${cls||''}" contenteditable="true">${md(item)}</span></li>`).join('')+'</ul>';
-}
-
-projects.forEach((proj, pi) => {
-  let h = `<div class="project"><div class="project-header">
-    <span class="project-title">${esc(proj.name)}</span>
-    <span class="project-domain">${esc(proj.domain)} / ${esc(proj.subdomain)}</span>
-  </div>`;
-
-  proj.skills.forEach((sk, si) => {
-    const id = `sk-${pi}-${si}`;
-    const cn = sk.category.split('-')[0];
-    h += `<div class="skill" id="${id}-w">
-      <div class="skill-top">
-        <input type="checkbox" class="skill-cb" id="${id}" checked onchange="tog('${id}')">
-        <span class="badge c${cn}">${esc(CAT[sk.category]||sk.category)}</span>
-        <span class="skill-title">${mkEditable(sk.title, 'ed-title')}</span>
-        <span class="skill-conf">${esc(sk.confidence)}</span>
-      </div>
-      <div class="skill-desc">${mkEditable(sk.description, 'ed-desc')}</div>
-      <div class="skill-actions">
-        <button class="expand-btn" onclick="exp('${id}')" style="margin:0">▶ Show details</button>
-        <button class="gh-btn" onclick="openIssue(${pi},${si})" disabled>Submit to GitHub</button>
-      </div>
-      <div class="detail" id="${id}-d">
-        <h4>Domain Knowledge</h4>${mkEditable(sk.domain_knowledge, 'ed-dk')}
-        <h4>Reasoning Protocol</h4>${mkEditableList(sk.reasoning_steps, 'ed-rs')}
-        <h4>Tools</h4>${mkEditableList(sk.tools, 'ed-tools')}
-        <h4>Common Pitfalls</h4>${mkEditableList(sk.pitfalls, 'ed-pit')}
-      </div>
-    </div>`;
-  });
-
-  h += '</div>';
-  root.innerHTML += h;
-});
-
-function tog(id){document.getElementById(id+'-w').classList.toggle('rejected',!document.getElementById(id).checked)}
-function exp(id){
-  const d=document.getElementById(id+'-d'), b=d.previousElementSibling;
-  const open=d.classList.toggle('open');
-  b.textContent=open?'▼ Hide details':'▶ Show details';
-}
-
-function readEditable(parent, cls){
-  const el=parent.querySelector('.'+cls);
-  return el?el.textContent.trim():'';
-}
-function readEditableList(parent, cls){
-  return Array.from(parent.querySelectorAll('.'+cls)).map(e=>e.textContent.trim()).filter(Boolean);
-}
-
-function buildIssueUrl(pi, si){
-  const proj = projects[pi];
-  const sk = proj.skills[si];
-  const id = `sk-${pi}-${si}`;
-  const wrap = document.getElementById(id+'-w');
-  const detail = document.getElementById(id+'-d');
-
-  const authorName = document.getElementById('author-name').value.trim() || 'Anonymous';
-  const authorInst = document.getElementById('author-inst').value.trim();
-  const authorRole = document.getElementById('author-role').value;
-  const author = authorName + (authorInst ? ' (' + authorInst + ')' : '');
-
-  // Read current (possibly edited) values from DOM
-  const title = readEditable(wrap, 'ed-title') || sk.title;
-  const desc = readEditable(wrap, 'ed-desc') || sk.description;
-  const dk = readEditable(detail, 'ed-dk') || sk.domain_knowledge || '';
-  const rs = readEditableList(detail, 'ed-rs');
-  const tools = readEditableList(detail, 'ed-tools');
-  const pit = readEditableList(detail, 'ed-pit');
-
-  const slug = title.toLowerCase().replace(/[^a-z0-9]+/g,'-').replace(/^-|-$/g,'');
-  const subLabel = [proj.domain, proj.subdomain].map(s=>s.split('-').map(w=>w.charAt(0).toUpperCase()+w.slice(1)).join(' ')).join(' / ');
-  const catLabel = CAT[sk.category] || sk.category;
-  const date = DATA.date || new Date().toISOString().slice(0,10);
-
-  const skillMd = `---
-name: ${slug}
-description: >
-  ${desc}
-domain: ${proj.domain}
-subdomain: ${proj.subdomain}
-category: ${sk.category}
-author: "${author}"
-expertise_level: intermediate
-tags: []
-dependencies: []
-version: 1.0.0
-status: draft
-reviewed_by: []
----
-
-## Purpose
-
-${desc}
-
-## Tools
-
-${tools.length ? tools.map(t=>'- **'+t.split(' — ')[0]+'**'+(t.includes(' — ')?': '+t.split(' — ').slice(1).join(' — '):'')).join('\n') : '- N/A'}
-
-## Domain Knowledge
-
-### Key Concepts
-
-${dk || 'N/A'}
-
-## Reasoning Protocol
-
-${rs.length ? rs.map((s,i)=>'Step '+(i+1)+': '+s).join('\n') : 'N/A'}
-
-## Common Pitfalls
-
-${pit.length ? pit.map(p=>'- '+p).join('\n') : '- N/A'}
-
-## References
-
-- Extracted by /researchskills-extract on ${date}
-- Source project: ${proj.name}`;
-
-  const params = new URLSearchParams({
-    template: '01-submit-skill.yml',
-    title: 'Skill Submission: ' + slug,
-    english_name: authorName,
-    role: authorRole || '',
-    affiliation: authorInst || '',
-    subdomain: subLabel,
-    category: sk.category + ' — ' + catLabel,
-    skill_content: skillMd,
-    generation_method: 'Generated by /researchskills-extract, then reviewed by me',
-    notes: 'Confidence: ' + sk.confidence
-  });
-
-  return 'https://github.com/ScienceIntelligence/ResearchSkills/issues/new?' + params.toString();
-}
-
-function openIssue(pi, si){
-  if(!consentGiven){ alert('Please agree to the submission terms first.'); return; }
-  window.open(buildIssueUrl(pi, si), '_blank');
-}
-
-function submitAll(){
-  if(!consentGiven){ alert('Please agree to the submission terms first.'); return; }
-  let count = 0;
-  projects.forEach((proj, pi) => {
-    proj.skills.forEach((sk, si) => {
-      const id = `sk-${pi}-${si}`;
-      if(!document.getElementById(id).checked) return;
-      window.open(buildIssueUrl(pi, si), '_blank');
-      count++;
-    });
-  });
-  if(count===0){alert('No skills selected.');return}
-  document.getElementById('submit-msg').textContent = '✓ Opened ' + count + ' GitHub issue tab(s). Review and submit each one.';
-  document.getElementById('output').style.display = 'block';
-}
-</script>
-</body>
-</html>
diff --git a/researchskills-extract/templates/skill-template.md b/researchskills-extract/templates/skill-template.md
deleted file mode 100644
index 1df7925..0000000
--- a/researchskills-extract/templates/skill-template.md
+++ /dev/null
@@ -1,82 +0,0 @@
----
-# REQUIRED FIELDS — fill in all of these before opening a PR
-name: your-skill-name                        # lowercase, hyphens only. This is the Claude Code invocation name.
-description: >                               # 1-2 sentences: when should this skill be used?
-  Describe what this skill does and when to invoke it.
-domain: physics                              # physics | mathematics | computer-science | quantitative-biology | statistics | eess | economics | quantitative-finance
-subdomain: ""                                # more specific area (e.g. quantum-physics)
-author: "Full Name (Institution)"            # e.g. "Dr. Albert Einstein (ETH Zürich Physics)"
-expertise_level: advanced                    # beginner | intermediate | advanced
-status: draft                                # leave as draft; reviewer will update after review
----
-
-<!--
-  INSTRUCTIONS FOR CONTRIBUTORS
-  ─────────────────────────────
-  1. Delete these comment blocks before submitting.
-  2. Fill in ALL required frontmatter fields above.
-  3. Complete each section below. "Required" sections must be present.
-  4. Run `python utils/tools/validate.py .agents/skills/<domain>/<subdomain>/<your-skill>.md` before opening a PR.
-  5. See utils/SKILL_SCHEMA.md for full field documentation.
-
-  NOTE: The primary contribution method is now decision trees via /researchskills-extract.
-  This template is for manually written skills only.
--->
-
-## Purpose
-<!-- REQUIRED -->
-<!-- One paragraph: what problem does this skill solve? When should it be invoked?
-     Example: "Use this skill when analyzing experimental data involving quantum entanglement.
-     It guides the AI through Bell inequality calculations and statistical interpretation." -->
-
-## Tools
-<!-- REQUIRED -->
-<!-- List the key tools, software, libraries, databases, or instruments used in this domain.
-     For each tool, briefly describe what it does and when to use it.
-     Example: "BLAST (sequence alignment) — use when comparing DNA/protein sequences against a database." -->
-
-- **[Tool Name]**: what it does, when to use it
-- **[Tool Name]**: what it does, when to use it
-
-## Domain Knowledge
-<!-- REQUIRED -->
-<!-- The core content: key concepts, definitions, equations, established facts.
-     Be precise and comprehensive — this is what the AI will draw on. -->
-
-### Key Concepts
-
-### Fundamental Equations / Principles
-
-### Important Results & Theorems
-
-## Reasoning Protocol
-<!-- REQUIRED -->
-<!-- Numbered steps guiding AI reasoning for this domain. -->
-
-Step 1: 
-Step 2: 
-Step 3: 
-
-## Common Pitfalls
-<!-- REQUIRED -->
-<!-- Mistakes, misconceptions, edge cases, and counterintuitive results to watch for. -->
-
-- 
-
-## Examples
-<!-- RECOMMENDED -->
-<!-- One or more worked examples demonstrating correct application. -->
-
-### Example 1: [Title]
-
-**Problem:**
-
-**Reasoning:**
-
-**Answer:**
-
-## References
-<!-- RECOMMENDED -->
-<!-- Key papers, textbooks, or resources. Use standard citation format. -->
-
-- 
diff --git a/researchskills-extract/tests/fixtures/engineering-session.jsonl b/researchskills-extract/tests/fixtures/engineering-session.jsonl
deleted file mode 100644
index 54c003d..0000000
--- a/researchskills-extract/tests/fixtures/engineering-session.jsonl
+++ /dev/null
@@ -1,4 +0,0 @@
-{"type":"user","message":{"role":"user","content":[{"type":"text","text":"Help me set up a Next.js project with Tailwind CSS and deploy it to Vercel."}]},"uuid":"b001","timestamp":"2026-03-16T09:00:00Z","sessionId":"test-eng-001","cwd":"/Users/dev/projects/my-website"}
-{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"I'll help you set up a Next.js project with Tailwind CSS. Let's start with creating the project..."}]},"uuid":"b002","timestamp":"2026-03-16T09:01:00Z","sessionId":"test-eng-001"}
-{"type":"user","message":{"role":"user","content":[{"type":"text","text":"Now add a dark mode toggle to the navbar."}]},"uuid":"b003","timestamp":"2026-03-16T09:15:00Z","sessionId":"test-eng-001"}
-{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"Sure, I'll add a dark mode toggle component..."}]},"uuid":"b004","timestamp":"2026-03-16T09:16:00Z","sessionId":"test-eng-001"}
diff --git a/researchskills-extract/tests/fixtures/minimal-session.jsonl b/researchskills-extract/tests/fixtures/minimal-session.jsonl
deleted file mode 100644
index f7dafcb..0000000
--- a/researchskills-extract/tests/fixtures/minimal-session.jsonl
+++ /dev/null
@@ -1,2 +0,0 @@
-{"type":"user","message":{"role":"user","content":[{"type":"text","text":"hello"}]},"uuid":"c001","timestamp":"2026-03-17T08:00:00Z","sessionId":"test-minimal-001","cwd":"/Users/scientist/projects/test"}
-{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"Hello! How can I help you today?"}]},"uuid":"c002","timestamp":"2026-03-17T08:00:05Z","sessionId":"test-minimal-001"}
diff --git a/researchskills-extract/tests/fixtures/research-session.jsonl b/researchskills-extract/tests/fixtures/research-session.jsonl
deleted file mode 100644
index 9453d09..0000000
--- a/researchskills-extract/tests/fixtures/research-session.jsonl
+++ /dev/null
@@ -1,6 +0,0 @@
-{"type":"user","message":{"role":"user","content":[{"type":"text","text":"I'm working on density functional theory calculations for lithium iron phosphate. I need to set up VASP input files with the correct INCAR parameters for a GGA+U calculation. The U value for Fe 3d orbitals should be around 4.3 eV based on the literature. Can you help me generate the INCAR file?"}]},"uuid":"a001","timestamp":"2026-03-15T10:00:00Z","sessionId":"test-research-001","cwd":"/Users/scientist/projects/lfp-dft"}
-{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"I'll help you set up the INCAR file for your GGA+U calculation on LiFePO4. Here are the key parameters..."}]},"uuid":"a002","timestamp":"2026-03-15T10:01:00Z","sessionId":"test-research-001"}
-{"type":"user","message":{"role":"user","content":[{"type":"text","text":"The convergence is not working. EDIFF is set to 1E-6 but after 200 ionic steps it still hasn't converged. I think the issue might be with the mixing parameters. In my experience, for transition metal oxides you need to use a smaller AMIX value, around 0.05 instead of the default 0.4. Also BMIX should be reduced. Let me try that."}]},"uuid":"a003","timestamp":"2026-03-15T10:15:00Z","sessionId":"test-research-001"}
-{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"Good observation. For transition metal systems with localized d-electrons, the default mixing parameters are indeed too aggressive..."}]},"uuid":"a004","timestamp":"2026-03-15T10:16:00Z","sessionId":"test-research-001"}
-{"type":"user","message":{"role":"user","content":[{"type":"text","text":"It converged! The key insight was reducing AMIX to 0.05 and BMIX to 0.0001 for the Fe d-orbital system. This is a pattern I've seen across multiple LFP and LMO calculations. We should document this as a general rule for GGA+U on transition metal oxides."}]},"uuid":"a005","timestamp":"2026-03-15T10:30:00Z","sessionId":"test-research-001"}
-{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"Excellent! This is indeed a well-known but under-documented pattern in the DFT community..."}]},"uuid":"a006","timestamp":"2026-03-15T10:31:00Z","sessionId":"test-research-001"}
diff --git a/researchskills-extract/tests/test-postinstall.js b/researchskills-extract/tests/test-postinstall.js
deleted file mode 100644
index f702374..0000000
--- a/researchskills-extract/tests/test-postinstall.js
+++ /dev/null
@@ -1,57 +0,0 @@
-#!/usr/bin/env node
-"use strict";
-
-const fs = require("fs");
-const path = require("path");
-const os = require("os");
-const { execFileSync } = require("child_process");
-
-const COMMANDS_DIR = path.join(os.homedir(), ".claude", "commands");
-const TARGET = path.join(COMMANDS_DIR, "researchskills-extract.md");
-const SCRIPT_DIR = path.join(__dirname, "..", "scripts");
-
-let passed = 0;
-let failed = 0;
-
-function assert(condition, msg) {
-  if (condition) {
-    console.log("  ✓ " + msg);
-    passed++;
-  } else {
-    console.error("  ✗ " + msg);
-    failed++;
-  }
-}
-
-// Clean up any existing file first
-if (fs.existsSync(TARGET)) {
-  fs.unlinkSync(TARGET);
-}
-
-console.log("Test: postinstall.js");
-execFileSync(process.execPath, [path.join(SCRIPT_DIR, "postinstall.js")], { stdio: "pipe" });
-assert(fs.existsSync(TARGET), "Command file exists after install");
-
-const content = fs.readFileSync(TARGET, "utf-8");
-assert(content.includes("researchskills-extract"), "Command file contains expected content");
-assert(content.startsWith("#"), "Command file starts with markdown header");
-
-console.log("\nTest: consent gate");
-const REPORT_TEMPLATE = path.join(__dirname, "..", "templates", "report.html");
-const reportTemplate = fs.readFileSync(REPORT_TEMPLATE, "utf-8");
-assert(reportTemplate.includes('id="consent-cb"'), "Report template has consent checkbox");
-assert(reportTemplate.includes('disabled'), "Submit buttons are disabled by default");
-assert(reportTemplate.includes('consentGiven'), "Report template checks consent before submission");
-
-console.log("\nTest: finalize.js requires --upload flag");
-const FINALIZE_SCRIPT = path.join(__dirname, "..", "scripts", "finalize.js");
-const finalizeContent = fs.readFileSync(FINALIZE_SCRIPT, "utf-8");
-assert(finalizeContent.includes("if (!options.upload)"), "finalize.js defaults to no-upload without --upload flag");
-assert(finalizeContent.includes("'--upload'"), "finalize.js supports --upload CLI flag");
-
-console.log("\nTest: postuninstall.js");
-execFileSync(process.execPath, [path.join(SCRIPT_DIR, "postuninstall.js")], { stdio: "pipe" });
-assert(!fs.existsSync(TARGET), "Command file removed after uninstall");
-
-console.log("\n" + passed + " passed, " + failed + " failed");
-process.exit(failed > 0 ? 1 : 0);
diff --git a/utils/CONTRIBUTING.md b/utils/CONTRIBUTING.md
index 1368e14..ce95ffc 100644
--- a/utils/CONTRIBUTING.md
+++ b/utils/CONTRIBUTING.md
@@ -2,8 +2,8 @@
 
 Thank you for contributing your expertise! There are three ways to contribute:
 
-1. **Extract from AI history (recommended):** Use `/researchskills-extract` to automatically extract research skills from your Claude Code or Codex conversation history.
-2. **Convert existing skills:** Use `/researchskills-convert` to convert skills you already have (notes, documents, any format) into ResearchSkills format and open a PR.
+1. **Extract from AI history (recommended):** Use the included ResearchSkills meta skill to extract research skills from your Claude Code or Codex conversation history.
+2. **Convert existing skills:** Use the same meta skill to convert skills you already have (notes, documents, any format) into ResearchSkills format and open a PR.
 3. **Write manually:** Write a skill file by hand following the templates.
 
 ---
@@ -12,14 +12,9 @@ Thank you for contributing your expertise! There are three ways to contribute:
 
 ### Via Claude Code / Codex
 
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
+Install the library with `npx skills add ScienceIntelligence/ResearchSkills` or `bunx skills add ScienceIntelligence/ResearchSkills`, then ask your agent to use `researchskills-meta-skill-extraction-and-conversion`.
 
-Then in Claude Code: `/researchskills-extract`
-Or in Codex: `$researchskills-extract`
-
-The tool scans your conversation history, extracts research skills (procedural, semantic, and episodic), and opens a browser review page. After reviewing, submit via GitHub.
+The meta skill scans the source material you provide, extracts research skills (procedural, semantic, and episodic), writes them under `.agents/skills/`, and prepares them for GitHub PR submission.
 
 ### Via Web (ChatGPT / Claude / Gemini)
 
@@ -29,16 +24,9 @@ https://researchskills.ai/submit-manually
 
 ## 2. Convert Existing Skills
 
-Already have research skills in notes, documents, or any format? Use `/researchskills-convert` to convert them and open a PR — all in one step.
-
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
-
-Then in Claude Code: `/researchskills-convert`
-Or in Codex: `$researchskills-convert`
+Already have research skills in notes, documents, local skills, prompts, rubrics, or any format? Use `researchskills-meta-skill-extraction-and-conversion` to convert them and open a PR.
 
-The command will ask where your skills are (file path or directory), read them, convert each one into the correct format, and open a PR to this repository. It handles forking, branching, file placement, and de-identification automatically.
+The meta skill asks where your source material is, reads it, converts each reusable research know-how item into the correct format, and places files under `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`. It should preserve scientific content while removing private identifiers.
 
 ---
 
@@ -124,8 +112,8 @@ Open a GitHub Discussion or reach out to the core team via issues.
 
 感谢你贡献专业知识！有三种贡献方式：
 
-1. **从 AI 历史中提取（推荐）：** 使用 `/researchskills-extract` 从 Claude Code 或 Codex 对话历史中自动提取科研技能。
-2. **转换已有技能：** 使用 `/researchskills-convert` 将你已有的技能（笔记、文档、任何格式）转换为 ResearchSkills 格式并自动提交 PR。
+1. **从 AI 历史中提取（推荐）：** 使用内置的 ResearchSkills meta skill，从 Claude Code 或 Codex 对话历史中自动提取科研技能。
+2. **转换已有技能：** 使用同一个 meta skill，将你已有的技能（笔记、文档、任何格式）转换为 ResearchSkills 格式并自动提交 PR。
 3. **手动撰写：** 参照模板手动撰写 Skill 文件。
 
 ---
@@ -134,14 +122,9 @@ Open a GitHub Discussion or reach out to the core team via issues.
 
 ### 通过 Claude Code / Codex
 
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
+先用 `npx skills add ScienceIntelligence/ResearchSkills` 或 `bunx skills add ScienceIntelligence/ResearchSkills` 安装技能库，然后让你的 agent 使用 `researchskills-meta-skill-extraction-and-conversion`。
 
-在 Claude Code 中运行: `/researchskills-extract`
-在 Codex 中运行: `$researchskills-extract`
-
-工具会扫描你的对话历史，提取科研技能（程序性、语义性和情景性），并在浏览器中打开审阅页面。审阅后通过 GitHub 提交。
+这个 meta skill 会扫描你提供的源材料，提取科研技能（程序性、语义性和情景性），写入 `.agents/skills/`，并准备通过 GitHub PR 提交。
 
 ### 通过网页版（ChatGPT / Claude / Gemini）
 
@@ -151,16 +134,9 @@ https://researchskills.ai/submit-manually
 
 ## 2. 转换已有技能
 
-已有科研技能记录——笔记、文档或任何格式？使用 `/researchskills-convert` 一键转换并提交 PR。
-
-```bash
-npm install -g @scienceintelligence/researchskills-extract
-```
-
-在 Claude Code 中运行: `/researchskills-convert`
-在 Codex 中运行: `$researchskills-convert`
+已有科研技能记录、笔记、文档、本地 skill、prompt 或 rubric？使用 `researchskills-meta-skill-extraction-and-conversion` 转换并提交 PR。
 
-该命令会询问你的技能文件位置（文件路径或目录），读取并转换为正确格式，然后自动向本仓库提交 PR。Fork、分支创建、文件放置和去标识化均自动完成。
+这个 meta skill 会询问源材料位置，读取并把每个可复用科研 know-how 转换为正确格式，放到 `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`。它应保留科学内容，同时移除私有标识。
 
 ---
 
diff --git a/utils/scripts/generate-taxonomy.js b/utils/scripts/generate-taxonomy.js
index 4528f6a..5020308 100644
--- a/utils/scripts/generate-taxonomy.js
+++ b/utils/scripts/generate-taxonomy.js
@@ -6,7 +6,7 @@
  * mapping domain → sorted list of subdomains. This is the single source of
  * truth for the arXiv-aligned taxonomy used by:
  *   - webserver review page (subdomain dropdown)
- *   - researchskills-extract CLI (domain mapping)
+ *   - ResearchSkills meta skill workflows (domain mapping)
  *   - any future tooling that needs the list
  *
  * Usage (from repo root):

From 4179c7e04ba5c2932cd892baaacd333c43561692 Mon Sep 17 00:00:00 2001
From: LIghtJUNction <lightjunction.me@gmail.com>
Date: Mon, 1 Jun 2026 04:59:17 +0800
Subject: [PATCH 3/4] Move researchskills extract to meta skill

---
 .../SKILL.md}                                 | 10 +++----
 .../02-submit-decision-tree.yml               |  6 ++---
 AGENT.md                                      |  2 +-
 README.md                                     |  2 +-
 readme_zh.md                                  |  2 +-
 utils/CONTRIBUTING.md                         |  8 +++---
 utils/tools/validate.py                       | 26 +++++++++++++++++--
 7 files changed, 37 insertions(+), 19 deletions(-)
 rename .agents/skills/{computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md => researchskills-extract/SKILL.md} (90%)

diff --git a/.agents/skills/computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md b/.agents/skills/researchskills-extract/SKILL.md
similarity index 90%
rename from .agents/skills/computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md
rename to .agents/skills/researchskills-extract/SKILL.md
index 008e674..fac2030 100644
--- a/.agents/skills/computer-science/artificial-intelligence/scienceintelligence/procedural/tie--researchskills-meta-skill-extraction-and-conversion.md
+++ b/.agents/skills/researchskills-extract/SKILL.md
@@ -1,14 +1,10 @@
 ---
-name: researchskills-meta-skill-extraction-and-conversion
-memory_type: procedural
-subtype: tie
-domain: computer-science
-subdomain: artificial-intelligence
-contributor: scienceintelligence
+name: researchskills-extract
+description: Extract or convert existing research know-how into ResearchSkills files under .agents/skills/.
 ---
 
 ## When
-Use this meta skill when the task is to create ResearchSkills from existing material rather than to apply a domain skill directly. Typical inputs include AI conversation history, local agent skills, slash commands, memory files, rubrics, prompts, lab notes, research logs, or project instructions that may contain reusable research know-how.
+Use this meta skill when the task is to create ResearchSkills from existing material rather than to apply a domain skill directly. Typical inputs include AI conversation history, local agent skills, memory files, rubrics, prompts, lab notes, research logs, or project instructions that may contain reusable research know-how.
 
 Do not use it for generic engineering, DevOps, UI, database, Docker, package-manager, git, or textbook material unless the content directly changes how research should be done.
 
diff --git a/.github/ISSUE_TEMPLATE/02-submit-decision-tree.yml b/.github/ISSUE_TEMPLATE/02-submit-decision-tree.yml
index 43aa5c2..63674c4 100644
--- a/.github/ISSUE_TEMPLATE/02-submit-decision-tree.yml
+++ b/.github/ISSUE_TEMPLATE/02-submit-decision-tree.yml
@@ -1,5 +1,5 @@
 name: Submit a Decision Tree
-description: Submit a research decision tree generated by /researchskills-extract or the web prompt
+description: Submit a research decision tree generated by the researchskills-extract meta skill or the web prompt
 title: "Decision Tree: [project or paper title]"
 labels: ["decision-tree-submission"]
 body:
@@ -9,7 +9,7 @@ body:
         Submit your research decision tree here! Paste the full JSON output below.
 
         **How to generate a decision tree:**
-        - **Claude Code / Codex:** Install `npm install -g @scienceintelligence/researchskills-extract`, then run `/researchskills-extract`
+        - **Claude Code / Codex:** Install `npx skills add ScienceIntelligence/ResearchSkills` or `bunx skills add ScienceIntelligence/ResearchSkills`, then ask your agent to use `researchskills-extract`
         - **Web (ChatGPT / Claude / Gemini):** Use the one-click prompt from the [README](https://github.com/ScienceIntelligence/ResearchSkills#method-b-one-click-prompt-for-web-users-chatgpt--claude--gemini)
 
   - type: input
@@ -247,7 +247,7 @@ body:
     attributes:
       label: How was this tree created?
       options:
-        - "Generated by /researchskills-extract, then reviewed by me"
+        - "Generated by researchskills-extract, then reviewed by me"
         - "Generated via web prompt, then reviewed by me"
         - "Constructed manually"
     validations:
diff --git a/AGENT.md b/AGENT.md
index 01994a6..398b733 100644
--- a/AGENT.md
+++ b/AGENT.md
@@ -3,7 +3,7 @@
 ## Repository Layout
 
 - Public research skills live under `.agents/skills/`.
-- The extraction/conversion workflow is a normal meta skill under `.agents/skills/`, not a separate package.
+- The extraction/conversion workflow is the normal meta skill at `.agents/skills/researchskills-extract/SKILL.md`, not a separate package or command.
 - Repository validators and contribution examples should point at `.agents/skills/`, not a top-level `skills/` directory.
 
 ## Editing Rules
diff --git a/README.md b/README.md
index 0766770..ee13c73 100644
--- a/README.md
+++ b/README.md
@@ -74,7 +74,7 @@ bunx skills add ScienceIntelligence/ResearchSkills
 
 ### Method A: Use the ResearchSkills Meta Skill (Recommended)
 
-Install the library, then ask Claude Code, Codex, or another markdown-aware agent to use the `researchskills-meta-skill-extraction-and-conversion` skill. This is a **meta skill**: it turns conversation history, local skills, prompts, rubrics, notes, or other research know-how into ResearchSkills files.
+Install the library, then ask Claude Code, Codex, or another markdown-aware agent to use the `researchskills-extract` skill. This is a **meta skill**: it turns conversation history, local skills, prompts, rubrics, notes, or other research know-how into ResearchSkills files.
 
 Example prompt:
 
diff --git a/readme_zh.md b/readme_zh.md
index 9f5aab9..bee4b02 100644
--- a/readme_zh.md
+++ b/readme_zh.md
@@ -72,7 +72,7 @@ bunx skills add ScienceIntelligence/ResearchSkills
 
 ### 方式 A：使用 ResearchSkills Meta Skill（推荐）
 
-安装技能库后，让 Claude Code、Codex 或任何能读取 markdown skill 的 agent 使用 `researchskills-meta-skill-extraction-and-conversion`。这是一个 **meta skill**：它把对话历史、本地 skill、prompt、rubric、笔记或其他科研 know-how 转成 ResearchSkills 文件。
+安装技能库后，让 Claude Code、Codex 或任何能读取 markdown skill 的 agent 使用 `researchskills-extract`。这是一个 **meta skill**：它把对话历史、本地 skill、prompt、rubric、笔记或其他科研 know-how 转成 ResearchSkills 文件。
 
 示例 prompt：
 
diff --git a/utils/CONTRIBUTING.md b/utils/CONTRIBUTING.md
index ce95ffc..4505347 100644
--- a/utils/CONTRIBUTING.md
+++ b/utils/CONTRIBUTING.md
@@ -12,7 +12,7 @@ Thank you for contributing your expertise! There are three ways to contribute:
 
 ### Via Claude Code / Codex
 
-Install the library with `npx skills add ScienceIntelligence/ResearchSkills` or `bunx skills add ScienceIntelligence/ResearchSkills`, then ask your agent to use `researchskills-meta-skill-extraction-and-conversion`.
+Install the library with `npx skills add ScienceIntelligence/ResearchSkills` or `bunx skills add ScienceIntelligence/ResearchSkills`, then ask your agent to use `researchskills-extract`.
 
 The meta skill scans the source material you provide, extracts research skills (procedural, semantic, and episodic), writes them under `.agents/skills/`, and prepares them for GitHub PR submission.
 
@@ -24,7 +24,7 @@ https://researchskills.ai/submit-manually
 
 ## 2. Convert Existing Skills
 
-Already have research skills in notes, documents, local skills, prompts, rubrics, or any format? Use `researchskills-meta-skill-extraction-and-conversion` to convert them and open a PR.
+Already have research skills in notes, documents, local skills, prompts, rubrics, or any format? Use `researchskills-extract` to convert them and open a PR.
 
 The meta skill asks where your source material is, reads it, converts each reusable research know-how item into the correct format, and places files under `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`. It should preserve scientific content while removing private identifiers.
 
@@ -122,7 +122,7 @@ Open a GitHub Discussion or reach out to the core team via issues.
 
 ### 通过 Claude Code / Codex
 
-先用 `npx skills add ScienceIntelligence/ResearchSkills` 或 `bunx skills add ScienceIntelligence/ResearchSkills` 安装技能库，然后让你的 agent 使用 `researchskills-meta-skill-extraction-and-conversion`。
+先用 `npx skills add ScienceIntelligence/ResearchSkills` 或 `bunx skills add ScienceIntelligence/ResearchSkills` 安装技能库，然后让你的 agent 使用 `researchskills-extract`。
 
 这个 meta skill 会扫描你提供的源材料，提取科研技能（程序性、语义性和情景性），写入 `.agents/skills/`，并准备通过 GitHub PR 提交。
 
@@ -134,7 +134,7 @@ https://researchskills.ai/submit-manually
 
 ## 2. 转换已有技能
 
-已有科研技能记录、笔记、文档、本地 skill、prompt 或 rubric？使用 `researchskills-meta-skill-extraction-and-conversion` 转换并提交 PR。
+已有科研技能记录、笔记、文档、本地 skill、prompt 或 rubric？使用 `researchskills-extract` 转换并提交 PR。
 
 这个 meta skill 会询问源材料位置，读取并把每个可复用科研 know-how 转换为正确格式，放到 `.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md`。它应保留科学内容，同时移除私有标识。
 
diff --git a/utils/tools/validate.py b/utils/tools/validate.py
index b9fb211..52e0c92 100755
--- a/utils/tools/validate.py
+++ b/utils/tools/validate.py
@@ -35,8 +35,13 @@
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent
 SKILLS_DIR = REPO_ROOT / ".agents" / "skills"
 
-# Derive valid domains from directory structure (single source of truth)
-VALID_DOMAINS = {p.name for p in SKILLS_DIR.iterdir() if p.is_dir() and not p.name.startswith('.')} if SKILLS_DIR.is_dir() else set()
+# Derive valid domains from directory structure (single source of truth).
+# Top-level meta skills use .agents/skills/<skill>/SKILL.md and are not domains.
+VALID_DOMAINS = {
+    p.name
+    for p in SKILLS_DIR.iterdir()
+    if p.is_dir() and not p.name.startswith(".") and not (p / "SKILL.md").exists()
+} if SKILLS_DIR.is_dir() else set()
 
 
 def slugify(value: str) -> str:
@@ -56,6 +61,9 @@ def validate_path(path: Path) -> list[str]:
         return errors
 
     parts = rel.parts  # e.g. ('physics', 'geophysics', 'jdoe', 'semantic', 'correction--x.md')
+    if len(parts) == 2 and parts[1] == "SKILL.md":
+        return errors
+
     if len(parts) < 5:
         errors.append(f"File must be in .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/, got: .agents/skills/{'/'.join(parts)}")
         return errors
@@ -105,6 +113,20 @@ def validate_file(path: Path) -> list[str]:
         errors.append("Frontmatter is not a YAML mapping")
         return errors
 
+    try:
+        rel = path.resolve().relative_to(SKILLS_DIR)
+    except ValueError:
+        rel = None
+
+    if rel and len(rel.parts) == 2 and rel.parts[1] == "SKILL.md":
+        if not isinstance(front.get("name"), str):
+            errors.append("Missing required field: 'name'")
+        if not isinstance(front.get("description"), str):
+            errors.append("Missing required field: 'description'")
+        if front.get("name") and front["name"] != rel.parts[0]:
+            errors.append(f"'name' field '{front['name']}' does not match skill folder '{rel.parts[0]}'")
+        return errors
+
     # Check required fields
     for field, ftype in REQUIRED_FIELDS.items():
         if field not in front or front[field] is None:

From 448f33412bcea65a6992f760f6fdb6913d681585 Mon Sep 17 00:00:00 2001
From: LIghtJUNction <lightjunction.me@gmail.com>
Date: Mon, 1 Jun 2026 05:10:59 +0800
Subject: [PATCH 4/4] Add helper scripts for researchskills extract

---
 .../skills/researchskills-extract/SKILL.md    |  26 ++-
 .../scripts/researchskills-extract.ps1        | 193 ++++++++++++++++++
 .../scripts/researchskills-extract.sh         | 185 +++++++++++++++++
 3 files changed, 403 insertions(+), 1 deletion(-)
 create mode 100644 .agents/skills/researchskills-extract/scripts/researchskills-extract.ps1
 create mode 100644 .agents/skills/researchskills-extract/scripts/researchskills-extract.sh

diff --git a/.agents/skills/researchskills-extract/SKILL.md b/.agents/skills/researchskills-extract/SKILL.md
index fac2030..a26a12e 100644
--- a/.agents/skills/researchskills-extract/SKILL.md
+++ b/.agents/skills/researchskills-extract/SKILL.md
@@ -13,7 +13,8 @@ Treat extraction and conversion as a normal ResearchSkills skill workflow, not a
 
 Preferred:
 - Ask for the source path, conversation export, or pasted source material.
-- Read only the requested source and directly referenced files needed to understand it.
+- Use `scripts/researchskills-extract.sh` on macOS/Linux or `scripts/researchskills-extract.ps1` on Windows when the source is a file or directory. The helper builds a bounded source bundle, a manifest, and an extraction prompt without installing anything globally.
+- Read only the requested source bundle and directly referenced files needed to understand it.
 - Extract one reusable research know-how item per output skill.
 - Assign exactly one memory type: `procedural`, `semantic`, or `episodic`.
 - Assign exactly one subtype:
@@ -24,6 +25,28 @@ Preferred:
 - Remove private paths, usernames, private URLs, project names, and collaborator names while preserving scientific parameters, methods, model names, datasets, and mechanisms.
 - Validate the generated files, then submit them by GitHub PR.
 
+Helper examples:
+
+```bash
+.agents/skills/researchskills-extract/scripts/researchskills-extract.sh path/to/source \
+  --domain computer-science \
+  --subdomain artificial-intelligence \
+  --contributor your-github-handle
+```
+
+```powershell
+.agents/skills/researchskills-extract/scripts/researchskills-extract.ps1 path\to\source `
+  -Domain computer-science `
+  -Subdomain artificial-intelligence `
+  -Contributor your-github-handle
+```
+
+After running the helper, read the generated `extraction-prompt.md`, create skill files under `.agents/skills/`, and validate them with:
+
+```bash
+python utils/tools/validate.py .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+```
+
 Rejected:
 - Do not install or document a separate npm package for this workflow.
 - Do not preserve legacy slash-command framing when a normal skill invocation is enough.
@@ -35,6 +58,7 @@ Reasoning: the extraction workflow is itself a meta skill: it teaches an agent h
 ## Local Verifiers
 - Every generated skill has frontmatter fields: `name`, `memory_type`, `subtype`, `domain`, `subdomain`, and `contributor`.
 - The file path matches its frontmatter memory type and subtype.
+- The helper output includes `source-manifest.tsv`, `source-bundle.md`, and `extraction-prompt.md`.
 - The body contains a concrete trigger, action or fact, reasoning mechanism, and failure or anti-example guidance when applicable.
 - Private identifiers are removed, but scientific content remains specific enough to be useful.
 - Repository validation passes for the generated skill files.
diff --git a/.agents/skills/researchskills-extract/scripts/researchskills-extract.ps1 b/.agents/skills/researchskills-extract/scripts/researchskills-extract.ps1
new file mode 100644
index 0000000..b3a9baa
--- /dev/null
+++ b/.agents/skills/researchskills-extract/scripts/researchskills-extract.ps1
@@ -0,0 +1,193 @@
+param(
+  [Parameter(Position = 0)]
+  [string]$Source,
+
+  [string]$Out,
+  [string]$Domain = "computer-science",
+  [string]$Subdomain = "artificial-intelligence",
+  [string]$Contributor = $env:USERNAME,
+  [int]$MaxBytes = 200000,
+  [string]$ValidateDir,
+  [switch]$Help
+)
+
+Set-StrictMode -Version Latest
+$ErrorActionPreference = "Stop"
+
+function Show-Usage {
+  @"
+Usage:
+  .\researchskills-extract.ps1 <source-file-or-dir> [options]
+
+Options:
+  -Out <dir>             Output directory for bundle/prompt.
+  -Domain <slug>         Target ResearchSkills domain.
+  -Subdomain <slug>      Target ResearchSkills subdomain.
+  -Contributor <handle>  Public contributor handle.
+  -MaxBytes <n>          Max bytes copied into source-bundle.md. Default: 200000.
+  -ValidateDir <dir>     Validate existing generated .md files under this repo path.
+  -Help                  Show this help.
+
+This helper does not call an LLM and does not install commands. It prepares a
+bounded source bundle and a high-signal extraction prompt for the agent.
+"@
+}
+
+if ($Help) {
+  Show-Usage
+  exit 0
+}
+
+$ScriptDir = Split-Path -Parent $MyInvocation.MyCommand.Path
+$RepoRoot = Resolve-Path (Join-Path $ScriptDir "../../../..")
+$Validator = Join-Path $RepoRoot "utils/tools/validate.py"
+
+if ($ValidateDir) {
+  if (-not (Test-Path $Validator)) {
+    throw "validator not found: $Validator"
+  }
+  $files = Get-ChildItem -Path $ValidateDir -Recurse -File -Filter "*.md" | ForEach-Object { $_.FullName }
+  if (-not $files -or $files.Count -eq 0) {
+    throw "no markdown files found under $ValidateDir"
+  }
+  & python $Validator @files
+}
+
+if (-not $Source) {
+  Show-Usage
+  exit 2
+}
+if (-not (Test-Path $Source)) {
+  throw "source not found: $Source"
+}
+if (-not $Contributor) {
+  $Contributor = "anonymous"
+}
+
+if (-not $Out) {
+  $stamp = Get-Date -Format "yyyyMMdd-HHmmss"
+  $Out = Join-Path $RepoRoot ".researchskills-extract/$stamp"
+}
+New-Item -ItemType Directory -Force -Path $Out | Out-Null
+
+$Manifest = Join-Path $Out "source-manifest.tsv"
+$Bundle = Join-Path $Out "source-bundle.md"
+$Prompt = Join-Path $Out "extraction-prompt.md"
+
+"path`tbytes" | Set-Content -Path $Manifest -Encoding utf8
+@"
+# ResearchSkills Source Bundle
+
+Source: $Source
+Generated: $((Get-Date).ToUniversalTime().ToString("yyyy-MM-ddTHH:mm:ssZ"))
+Byte limit: $MaxBytes
+
+"@ | Set-Content -Path $Bundle -Encoding utf8
+
+function Test-TextCandidate([string]$Path) {
+  $ext = [IO.Path]::GetExtension($Path).ToLowerInvariant()
+  return @(".md", ".markdown", ".txt", ".json", ".jsonl", ".yaml", ".yml", ".csv", ".tsv", ".log").Contains($ext)
+}
+
+function Append-SourceFile([string]$Path) {
+  if (-not (Test-Path $Path -PathType Leaf)) { return }
+  if (-not (Test-TextCandidate $Path)) { return }
+
+  $size = (Get-Item $Path).Length
+  $current = (Get-Item $Bundle).Length
+  if (($current + $size) -gt $MaxBytes) {
+    "$Path`t$size`tSKIPPED_LIMIT" | Add-Content -Path $Manifest -Encoding utf8
+    return
+  }
+
+  "$Path`t$size" | Add-Content -Path $Manifest -Encoding utf8
+  $text = Get-Content -Raw -Path $Path
+  $text = $text -replace "```", "` ` `"
+  @"
+
+
+## File: $Path
+
+``````text
+$text
+``````
+"@ | Add-Content -Path $Bundle -Encoding utf8
+}
+
+if (Test-Path $Source -PathType Leaf) {
+  Append-SourceFile (Resolve-Path $Source)
+} else {
+  $skipDirs = @(".git", "node_modules", ".venv", "venv", "dist", "build", "__pycache__", ".researchskills-extract")
+  Get-ChildItem -Path $Source -Recurse -File |
+    Where-Object {
+      $parts = $_.FullName -split [regex]::Escape([IO.Path]::DirectorySeparatorChar)
+      -not ($parts | Where-Object { $skipDirs -contains $_ })
+    } |
+    Sort-Object FullName |
+    ForEach-Object { Append-SourceFile $_.FullName }
+}
+
+$bundleText = Get-Content -Raw -Path $Bundle
+@"
+# ResearchSkills Extraction Prompt
+
+Use the ResearchSkills meta skill to extract reusable research know-how from
+the source bundle below. Write output files directly under:
+
+``.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md``
+
+Default metadata:
+
+- domain: ``$Domain``
+- subdomain: ``$Subdomain``
+- contributor: ``$Contributor``
+
+Quality gate:
+
+1. Extract only research know-how: scientific methodology, domain facts,
+   computational science methods, research design, evaluation strategy, or
+   concrete scientific episodes.
+2. Reject pure engineering, package-manager, git, auth, deployment, UI, or
+   generic coding lessons unless they directly constrain the research method.
+3. One skill file equals one reusable insight.
+4. De-identify private paths, usernames, project names, private URLs, and
+   collaborator names. Preserve scientific methods, model names, datasets,
+   parameters, and mechanisms.
+5. Prefer 0 high-quality skills over broad low-value skills.
+
+Memory types:
+
+- procedural: ``tie``, ``no-change``, ``constraint-failure``, ``operator-fail``
+- semantic: ``frontier``, ``non-public``, ``correction``
+- episodic: ``failure``, ``adaptation``, ``anomalous``
+
+Validation:
+
+``````bash
+python utils/tools/validate.py .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+``````
+
+After extraction, summarize:
+
+- files created
+- candidates rejected and why
+- validation command/result
+
+---
+
+$bundleText
+"@ | Set-Content -Path $Prompt -Encoding utf8
+
+@"
+ResearchSkills extraction helper complete.
+
+Output directory: $Out
+Manifest:         $Manifest
+Source bundle:    $Bundle
+Prompt:           $Prompt
+
+Next:
+  1. Read $Prompt.
+  2. Create ResearchSkills files under .agents/skills/.
+  3. Validate with: python utils/tools/validate.py <generated-skill.md>
+"@
diff --git a/.agents/skills/researchskills-extract/scripts/researchskills-extract.sh b/.agents/skills/researchskills-extract/scripts/researchskills-extract.sh
new file mode 100644
index 0000000..dcacfb4
--- /dev/null
+++ b/.agents/skills/researchskills-extract/scripts/researchskills-extract.sh
@@ -0,0 +1,185 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+usage() {
+  cat <<'USAGE'
+Usage:
+  researchskills-extract.sh <source-file-or-dir> [options]
+
+Options:
+  --out <dir>             Output directory for bundle/prompt.
+  --domain <slug>         Target ResearchSkills domain.
+  --subdomain <slug>      Target ResearchSkills subdomain.
+  --contributor <handle>  Public contributor handle.
+  --max-bytes <n>         Max bytes copied into source-bundle.md. Default: 200000.
+  --validate-dir <dir>    Validate existing generated .md files under this repo path.
+  -h, --help              Show this help.
+
+This helper does not call an LLM and does not install commands. It prepares a
+bounded source bundle and a high-signal extraction prompt for the agent.
+USAGE
+}
+
+die() {
+  printf 'ERROR: %s\n' "$*" >&2
+  exit 1
+}
+
+SOURCE=""
+OUT_DIR=""
+DOMAIN="computer-science"
+SUBDOMAIN="artificial-intelligence"
+CONTRIBUTOR="${USER:-anonymous}"
+MAX_BYTES="200000"
+VALIDATE_DIR=""
+
+while [ "$#" -gt 0 ]; do
+  case "$1" in
+    --out) OUT_DIR="${2:-}"; shift 2 ;;
+    --domain) DOMAIN="${2:-}"; shift 2 ;;
+    --subdomain) SUBDOMAIN="${2:-}"; shift 2 ;;
+    --contributor) CONTRIBUTOR="${2:-}"; shift 2 ;;
+    --max-bytes) MAX_BYTES="${2:-}"; shift 2 ;;
+    --validate-dir) VALIDATE_DIR="${2:-}"; shift 2 ;;
+    -h|--help) usage; exit 0 ;;
+    --*) die "unknown option: $1" ;;
+    *) if [ -z "$SOURCE" ]; then SOURCE="$1"; shift; else die "unexpected argument: $1"; fi ;;
+  esac
+done
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/../../../.." && pwd)"
+VALIDATOR="$REPO_ROOT/utils/tools/validate.py"
+
+if [ -n "$VALIDATE_DIR" ]; then
+  [ -f "$VALIDATOR" ] || die "validator not found: $VALIDATOR"
+  mapfile -d '' files < <(find "$VALIDATE_DIR" -type f -name '*.md' -print0)
+  [ "${#files[@]}" -gt 0 ] || die "no markdown files found under $VALIDATE_DIR"
+  python "$VALIDATOR" "${files[@]}"
+fi
+
+[ -n "$SOURCE" ] || { usage; exit 2; }
+[ -e "$SOURCE" ] || die "source not found: $SOURCE"
+case "$MAX_BYTES" in ''|*[!0-9]*) die "--max-bytes must be an integer" ;; esac
+
+if [ -z "$OUT_DIR" ]; then
+  stamp="$(date +%Y%m%d-%H%M%S)"
+  OUT_DIR="$REPO_ROOT/.researchskills-extract/$stamp"
+fi
+mkdir -p "$OUT_DIR"
+
+MANIFEST="$OUT_DIR/source-manifest.tsv"
+BUNDLE="$OUT_DIR/source-bundle.md"
+PROMPT="$OUT_DIR/extraction-prompt.md"
+
+printf 'path\tbytes\n' > "$MANIFEST"
+cat > "$BUNDLE" <<EOF
+# ResearchSkills Source Bundle
+
+Source: $SOURCE
+Generated: $(date -u +%Y-%m-%dT%H:%M:%SZ)
+Byte limit: $MAX_BYTES
+
+EOF
+
+is_text_candidate() {
+  case "$1" in
+    *.md|*.markdown|*.txt|*.json|*.jsonl|*.yaml|*.yml|*.csv|*.tsv|*.log) return 0 ;;
+    *) return 1 ;;
+  esac
+}
+
+append_file() {
+  file="$1"
+  [ -f "$file" ] || return 0
+  is_text_candidate "$file" || return 0
+  size="$(wc -c < "$file" | tr -d ' ')"
+  current="$(wc -c < "$BUNDLE" | tr -d ' ')"
+  if [ $((current + size)) -gt "$MAX_BYTES" ]; then
+    printf '%s\t%s\tSKIPPED_LIMIT\n' "$file" "$size" >> "$MANIFEST"
+    return 0
+  fi
+  printf '%s\t%s\n' "$file" "$size" >> "$MANIFEST"
+  {
+    printf '\n\n## File: %s\n\n' "$file"
+    printf '```text\n'
+    sed 's/```/` ` `/g' "$file"
+    printf '\n```\n'
+  } >> "$BUNDLE"
+}
+
+if [ -f "$SOURCE" ]; then
+  append_file "$SOURCE"
+else
+  while IFS= read -r -d '' file; do
+    append_file "$file"
+  done < <(
+    find "$SOURCE" \
+      \( -name .git -o -name node_modules -o -name .venv -o -name venv -o -name dist -o -name build -o -name __pycache__ -o -name .researchskills-extract \) -prune \
+      -o -type f -print0 | sort -z
+  )
+fi
+
+cat > "$PROMPT" <<EOF
+# ResearchSkills Extraction Prompt
+
+Use the ResearchSkills meta skill to extract reusable research know-how from
+the source bundle below. Write output files directly under:
+
+\`.agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md\`
+
+Default metadata:
+
+- domain: \`$DOMAIN\`
+- subdomain: \`$SUBDOMAIN\`
+- contributor: \`$CONTRIBUTOR\`
+
+Quality gate:
+
+1. Extract only research know-how: scientific methodology, domain facts,
+   computational science methods, research design, evaluation strategy, or
+   concrete scientific episodes.
+2. Reject pure engineering, package-manager, git, auth, deployment, UI, or
+   generic coding lessons unless they directly constrain the research method.
+3. One skill file equals one reusable insight.
+4. De-identify private paths, usernames, project names, private URLs, and
+   collaborator names. Preserve scientific methods, model names, datasets,
+   parameters, and mechanisms.
+5. Prefer 0 high-quality skills over broad low-value skills.
+
+Memory types:
+
+- procedural: \`tie\`, \`no-change\`, \`constraint-failure\`, \`operator-fail\`
+- semantic: \`frontier\`, \`non-public\`, \`correction\`
+- episodic: \`failure\`, \`adaptation\`, \`anomalous\`
+
+Validation:
+
+\`\`\`bash
+python utils/tools/validate.py .agents/skills/<domain>/<subdomain>/<contributor>/<memory_type>/<subtype>--<skill-name>.md
+\`\`\`
+
+After extraction, summarize:
+
+- files created
+- candidates rejected and why
+- validation command/result
+
+---
+
+$(cat "$BUNDLE")
+EOF
+
+cat <<EOF
+ResearchSkills extraction helper complete.
+
+Output directory: $OUT_DIR
+Manifest:         $MANIFEST
+Source bundle:    $BUNDLE
+Prompt:           $PROMPT
+
+Next:
+  1. Read $PROMPT.
+  2. Create ResearchSkills files under .agents/skills/.
+  3. Validate with: python utils/tools/validate.py <generated-skill.md>
+EOF