v0.7.0 — Unify & Complete: contract unification + helpers for the last 2 skills + full eval coverage#128
Merged
Merged
Conversation
…ast 2 skills, full eval coverage A major iteration closing the last structural gaps; three tracks, each locked by a gate. All on-philosophy (deterministic, progressive disclosure, no LLM-judgment gates, no pricing). Track A — Output-Contract unification: - Standardized all 13 diagnostic skills to the canonical Summary / Key Evidence / Remediation (+ Falsify / Risks floor); triage keeps Routing / Evidence Gaps; evidence-reporting aligned. Folded redundant root-cause-type/subsystem/failure-point lines into the single Primary Diagnosis line. Fixes the skill<->benchmark section divergence (a faithful report previously missed graded sections). - New gate scripts/contract_check.py (CI + unit test) enforces the vocabulary. Track B — complete the deterministic-helper principle: - triage: evidence_completeness_checker.py (present/missing + readiness vs required-evidence.md). - evidence-reporting: report_contract_validator.py (sections, confidence, redaction, no destructive recommendations). Both wired into SKILL.md + unit-tested. Every skill now has a deterministic helper. Track C — complete eval coverage: - Authored the 15 missing baselines (8 diagnosis + 7 routing) -> all 48 golden cases have a reference output (corpus exercised 100%, was ~68%). - Added a 3rd replication-versioning case (replication-dest-versioning-disabled). - New gate scripts/coverage_check.py (CI + unit test): every baseline-enabled category keeps >=2 cases and >=1 baseline. Validation: 11 gates green; 235 pytest (+16) + 21 extension tests; 48/48 baselines 100% PASS. Version 0.6.8 -> 0.7.0. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QhkyBLeDtBTmvTdVBqgCAV
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A major iteration that closes the last structural gaps instead of adding features. Three tracks, all on-philosophy (deterministic, progressive disclosure, no LLM-judgment gates, no pricing), each locked by a new gate.
Track A — Output-Contract unification
The eval corpus grades reports by section heading (
Summary/Key Evidence/Remediation), but everySKILL.mdOutput Contract used its own divergent vocabulary (Evidence,Recommendations,Fix,Root Cause, …) — so a model that faithfully followed a skill emitted sections the benchmark didn't credit. The skills and the benchmark disagreed on the output format.Summary / Key Evidence / Remediation(+ the existingWhat Would Falsify This/Risks / Open Questionsfloor); triage keepsRouting / Evidence Gaps; evidence-reporting aligned.Root cause type:/Subsystem:/Failure point:) into the singlePrimary Diagnosis: root_cause_type=…, affected_layer=…line.contract_check.py(CI + unit test) enforces the vocabulary so it can't drift again.Track B — complete the deterministic-helper principle
The two previously script-less skills get their first helpers (both unit-tested, wired into SKILL.md, provider-agnostic):
evidence_completeness_checker.py: deterministic present/missing + readiness score againstrequired-evidence.md— turns the "enough evidence?" judgment into a structural check.report_contract_validator.py: deterministic Output-Contract check (required sections, well-formed confidence, credential redaction, no destructive recommendations) — the same rules the eval applies, available before delivery.Every skill now has at least one deterministic helper.
Track C — complete eval coverage
replication-versioningcase (replication-dest-versioning-disabled) — the thinnest data-loss-adjacent skill.coverage_check.py(CI + unit test): every baseline-enabled category must keep ≥2 cases and ≥1 baseline reference.Validation
11 gates green (incl. the 2 new); 235 pytest (+16) + 21 extension tests; 48/48 baselines 100% PASS; routing-corpus intact. Version 0.6.8 → 0.7.0.
🤖 Generated with Claude Code
Generated by Claude Code