feat: Cortex v4.0.0 — signal-first, zero-decision#80
Merged
Conversation
…ision) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…inistic promotion, command consolidation - observe.py: tool output capture (10k cap) + real err_msg line + anti-false-positive guards - session-learner.js: draft/confirmed states (5 occ / 3 sessions), tool-name stopwords, action quality guards, per-proposal project attribution - distill_engine.py: deterministic law promotion (conf>=0.95 + 3 projects + occurrences_v4>=10), source instinct auto-archive, trigger prose derivation (no raw regex in laws), candidates mailbox removed, error-recovery moved to human validation - injector-engine.js: hollow-JSON + prompt-injection guards, draft filter, subtopic dedup, degenerate-trigger skip - commands: /cx-maintain (deterministic, cron-able) + /cx-review (single human digest) + cumulative /cx-eod; 17 legacy commands deprecated to stubs - session-start.py: review badge, EOD next-day Eisenhower matrix, maintain reminder Contract: docs/DESIGN-V4.md + docs/SPEC-PORT-SINAPSIS.md Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… docs & version quad-sync 4.0.0 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…raft status, codex guard, summaries archive Adversarial review (Codex GPT-5.5) found 2 critical + 3 major + 1 minor: - occurrences_v4 now incremented by the single YAML writer (session-learner) - _count_distinct_projects merges YAML + instinct-tracking + filesystem - auto-validated proposals born status:draft (5occ/3sessions contract holds) - strong error tokens override log-prefix guards ([codex] error: ... detected) - daily-summaries rotation archives instead of deleting - cx-status reads last_seen with lastSeen fallback Suite: 38/38 files, 693 passed / 0 failed. Live reinstall verified. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…gnored Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
1 participant
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.
Why
Three months of data showed the learning pipeline structurally frozen: last real law and skill promoted 2026-05-16, 88.5% historical proposal rejection, new instincts/month collapsing 43→15→6→0. Root causes (full forensics in
docs/audit-cortex-2026-07-02.mdand design rationale indocs/DESIGN-V4.md):observe.pyrarely populated output/err_msg, so error-fix proposals were garbage (grep headers and npm listings mistaken for errors).cx-*commands could not be scheduled; maintenance depended on memory and, when run, reported "nothing to do" because the signal had died upstream.law_eligible: true) that no automated path ever set: 0/71 instincts had it while 35 sat above the confidence threshold for 42-66 days.What
err_msgline extraction, word-boundary error patterns + anti-false-positive guards ([codex]/npm prefixes, version listings, grep headers) with strong-error-token override.status: draft(tracked, never injected) →confirmedonly at ≥5 occurrences across ≥3 sessions. Single-evidence auto-accept removed;error-recoverymoved to human validation.occurrences_v4≥10 + no recent noise. Source instinct auto-archived after promotion. Laws derive prose, never raw trigger regex (fixes the 40-char truncation corruption).law_eligible:falseremains an explicit veto; the manual:truerequirement and theauto-distill-candidates.mdmailbox are gone./cx-maintain(deterministic, cron-able: decay+dedup+purge+promotion+rotation+reconciliation+health),/cx-review(the ONLY human command: one weekly shorthand digest),/cx-eod(cumulative 24h, next-day Eisenhower block in SessionStart),/cx-status,/cx-gotcha,/cx-backup,/cx-restore. 17 legacy commands become deprecation stubs (docs/MIGRATION-V4.md).Tool|pattern) static skip, byte-stable ordering.resetAt; daily-summaries rotation archives instead of deleting.Verification
# retired in v4)./cx-maintainexecuted twice against real data: idempotent, produced a 21-item review digest, updated markers.🤖 Generated with Claude Code