skills-integrity: residuals batch (F12, F14-residual, F26) + mode-card integrity paragraphs

[aarontrowbridge]

Four-item residuals slice from the skills-integrity deep pass (audit-20260913-skills-deep-pass), per the director's brief. Draft — do not merge.

Item 1 (F12) — skills/research/SKILL.md

One reconciliation sentence at the campaign-ledger path definition: the session ledger lives in the personal vault (sessions/session-<YYYYMMDD>-<slug>.md, per director-core's discovery rule); project-local ledger/campaigns/ serves Research-Project-scoped work and reconciles to it at each loop boundary. Nothing else restructured.

Item 2 (F14-residual) — AGENTS.md

Setup clone URL corrected: harmoniqs/amicode (was the amicissimode typo).

Item 3 (F26) — vault-schemas/generate.mjs + committed contract

The LEGACY block mirrored the OLD amico-schema-check table. PR #1052 made amico-vault the single owner and explicitly flagged this generator as the follow-up slice ("the single-owner cleanup should extend to it as a separate slice; vault_schemas.test.ts pins those committed schemas against fixtures, so it belongs with its own RED→GREEN cycle") — this is that slice.

• Required sets now mirror the aligned amico-schema-check table (itself the field-for-field mirror of amico-vault): phantom required source dropped on insight/hypothesis/method; required-but-nullable session_id everywhere; paper gains title/date_read/relevance/systems; spec/plan gain visibility (plan also spec); retrospective gains outcome; org gains domain+relationship; device gains device_class+location; meeting gains org+topic.
• amico-vault's closed vocabularies are enums (task_type, statuses, relevance, outcome, failure_mode, device_class, relationship, visibility, ingest source); fidelity ∈ [0,1] and duration_us ≥ 0 bounds unchanged.
• Regenerated the 12 committed card schemas (node vault-schemas/generate.mjs); memory families, tension, tombstone, record schema, and plane-residency are untouched.
• Fixtures + vault_schemas.test.ts realigned in the same mechanical pass: legacy extras stay tolerated (additionalProperties: true), each invalid fixture keeps its intended violation as the only one, missing-required.json now pins $.session_id. 137/137 vault_schemas tests green (34 RED → 0 after alignment).

Item 4 — mode-card integrity paragraphs (ADR-0005 hold lifted)

≤3 lines per card, mode-binding sections only; DIRECTOR-SPINE and ledger-discovery-rule generated regions untouched; frontmatter untouched; agents/ mirrors modes/ byte-identically (AC9).

• develop — Integrate bullet: gate suite includes the skills-integrity lint when present on the branch; merged API-surface changes grep-checked against skill references (hits → ledger findings). Dispatch-discipline paragraph: loop-boundary ledger update carries the campaign's skill delta (findings filed, skills touched, proposals pending).
• research — loop-binding paragraph: analyzer cast carries the skill-friction check (instructions-vs-reality mismatches → findings; repeated un-skilled idioms → skill proposals); record boundary carries the same skill-delta row.

Verification — full pnpm --filter amicode test (fresh worktree, this branch)

	failed	passed	skipped
pristine origin/main baseline (same machine, separate worktree)	9	3468	49
this branch	10	3468	48

The 10 on this branch = the same 9 environmental baseline failures (amicode_service_runner ×1 app-dist, chat_bridge ×2 dev-tools symlink, skill_drift_lint ×2 Strumento fleet packages, main_source_resolver ×2, rebuild_coordinator ×2 timeouts) + cli_gate "really staged bins" ×1, which fails identically on pristine origin/main when run in isolation (it was skipped in the baseline full run — the local bin/dist staging artifact materialized between runs; 49th skip vs 48th). No failure is attributable to this diff.

Focused suites: vault_schemas 137/137; mode_cards, mode_cards_staging, mode_registry, mode_registry_staging, mode_block, mode_transition_engine, skills_integrity (real-tree lint green against the known list), workflow_skills_public — all green (191 passed, 1 pre-existing skip).

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 925eed41-4c65-48a6-a667-81ff1212c516

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}