Memo 097 rollout Phase A.
Remove the 33 per-area evaluator SKILL.md (11 areas x start-grade/evaluate/apply-improvement). The composed per-area prompt now comes from prompts.json.areas[] (PromptBuilder.build, PA-1) — the harness only passes ${area}-start-grade as a label to the injected skillInvoker; no fs lookup into skills/ existed on the runtime path.
- Knowledge nuggets already extracted to docs/improvement-loop.md + docs/harness.md (pointers updated to prompts.json.areas[]).
- Rewrote 2 coupled integration tests (skills.test.mjs, selection-aggregate-area.test.mjs) to drop SKILL.md-file assertions; preserved fixture<->output-schema conformance, HTTP-4xx-never-PASS, persona distribution. -117 SKILL.md-coupled assertions.
- Removed 'skills' from lint SCAN_DIRS.
- Did NOT touch provider domain skills (providers///skills/) or test fixtures.
Verify: 627/627 tests green; 0 SKILL.md remain; lint memo-refs 23->20 (pre-existing repo-wide lint:hygiene failure tracked separately).
NOTE: repo-wide npm run lint:hygiene is RED at HEAD (20 pre-existing internal memo-refs in provider-proof-sweep.mjs, ProviderProof.test.mjs, improvement-loop.md, etc. from earlier PRDs) — needs its own cleanup issue.
Memo 097 rollout Phase A.
Remove the 33 per-area evaluator SKILL.md (11 areas x start-grade/evaluate/apply-improvement). The composed per-area prompt now comes from
prompts.json.areas[](PromptBuilder.build, PA-1) — the harness only passes${area}-start-gradeas a label to the injected skillInvoker; no fs lookup into skills/ existed on the runtime path.Verify: 627/627 tests green; 0 SKILL.md remain; lint memo-refs 23->20 (pre-existing repo-wide lint:hygiene failure tracked separately).
NOTE: repo-wide
npm run lint:hygieneis RED at HEAD (20 pre-existing internal memo-refs in provider-proof-sweep.mjs, ProviderProof.test.mjs, improvement-loop.md, etc. from earlier PRDs) — needs its own cleanup issue.