Commit Graph

6 Commits

Author SHA1 Message Date
Boshu2 8bf08b1745 fix(flywheel): add executable lifecycle proof 2026-03-22 16:04:52 -04:00
Boshu2 961c9c5493 refactor(goals): reimagine GOALS.md from first principles + fix failing gates
Rewrote GOALS.md to align with product identity instead of tactical evolve
debt. New directives target untested product promises (multi-runtime, install
path, quarantined E2E tests, knowledge lifecycle). Retired vanity-metric
gates (coverage floors). Updated PRODUCT.md to reflect what the product
actually is today (three layers: skills, CLI, hooks).

Fixed 3 failing gates:
- checkSkills CC 24→<20 (extracted skillInstallDirs, scanSkillDir, skillOverlapWarning)
- parseFindingFile CC 22→<20 (extracted parseFindingFrontmatterFields, applyFindingField, parseFindingTitle)
- proof-run.sh: removed nonexistent --apply-decay flag from ao lookup

Fitness: 82.4% → 96.8% (15/16 passing, release-cadence transient)
2026-03-15 20:58:46 -04:00
Boden Fuller 4abd2a69be docs(context): sweep remaining ao inject references across docs/tests/scripts
Wave 3 of context-enforcement-phase3: replace ao inject workflow
guidance with ao lookup across documentation, test fixtures, and
scripts. Remaining 32 references are legitimate (inject command
implementation, CHANGELOG history, CLI reference tables, test
fixtures for command detection).

Files updated:
- 11 core docs (context-packet, seed-definition, leverage-points, etc.)
- 7 workflow diagram docs (brief, system-map, cli-skills-map, HOOKS, etc.)
- 4 scripts (proof-run, test-flywheel, check-skill-flag-refs, etc.)
- 3 shell tests (proof-run e2e, token-budgets, no-compaction)
- 1 Go test (json_validity_test.go)
- Production code: metrics_nudge.go, metrics_flywheel.go, hooks.go
2026-03-07 11:59:18 -05:00
Boden Fuller bcfdf74f47 evolve: cycle 24 -- flywheel-proof improved 2026-03-06 21:36:21 -05:00
Boden Fuller 97a3a4175f feat(context-leverage): resolve 9 post-mortem tech-debt items with TDD
Wave 1 (code):
- Extract formatVerdicts + renderHandoffField helpers (CC 25→19)
- Add maxGateRetryDepth=5 hard ceiling for gate retry recursion
- Remove dead selectHandoffFields (replaced by fieldAllowed)
- Replace bc dependency with awk in proof-run.sh
- Surface audit trail write errors via VerbosePrintf
- Consolidate hardcoded *4 magic constant → InjectCharsPerToken

Wave 2 (docs/skills):
- Document NarrativeCap=0 disambiguation semantics
- Add mandatory file-conflict matrix to /plan Step 5
- Add grep-for-existing-functions to /implement and /crank

11 new tests, all existing tests pass. Vibe: WARN (addressed).
2026-03-02 18:00:50 -05:00
Boden Fuller f9ce58d26f feat(context-leverage): wave 2 — adaptive assembly, token budget, flywheel proof
Track B: Phase context manifests with selective field loading. Each phase
declares what handoff fields it needs — Phase 2 gets decisions/risks,
Phase 3 gets artifacts. fieldAllowed() enables least-privilege context.

Track D: Mechanical token budget enforcement in buildHandoffContext.
estimateTokens (char/4), truncateToTokenBudget (sentence boundary),
applyContextBudget wired into manifest.MaxTokens.

Track G: Flywheel proof-run script + 3 Go integration tests verifying
cross-session knowledge retrieval, quality gate filtering, and decay.
2026-03-02 16:57:44 -05:00