docs(audit): refresh AUDIT-CHECKLIST/AUDIT-PROMPT post-C-1 + packages/ layout

[artyhoo]

What

Refresh the recursive self-audit tooling (AUDIT-CHECKLIST.md, AUDIT-PROMPT.md) to current origin/main reality. Primary driver: the merged C-1 agent-collision resolution (docs/meta-factory/research-patches/2026-05-20-agent-collision-resolution.md, PRs #79/#82/#83/#84). Internal-only tooling — not shipped via install.sh, no consumer impact.

⚠️ Scope note (read first)

The kickoff framed this as C-1-only. While reworking I found the audit files predate the packages/ monorepo migration (last touched at bb8c316, a factory/ rename) — so they referenced factory/RULES.md, scripts/audit-ai-docs.sh, tests/audit/ etc. that had moved. The C-1 probes are entangled with those dead paths (probe 2.1 / Phase-2 item 9 grep factory/RULES.md and best-practices-sidecar.md), so I could not "run the reworked probes" (T2) or satisfy §4.3 ("run end-to-end, no stale-probe artifacts") without also fixing the layout paths.

I reconciled both — C-1 (headline) + layout — because it's all in the same two files / one deliverable, and §4.3 mandates a runnable end-to-end audit. If you wanted strictly C-1, say so and I'll split the layout part out.

Per-reference classification

Reference (old)	Class	Now points at / asserts
best-practices-sidecar.md "validates RULES.md" / "manual review" (§2.1, §3.2, Phase-2 item 9)	REWORK	enforcement strategy = audit-script header (delegated-ESLint / probe / manual, lines 8-18) + aif-rules-check skill-context residue (R4/R10/R17). KEEP-AIF: not ours.
best-practices-sidecar / review-sidecar / docs-auditor boundaries (§3.4, item 13)	REWORK	the 3 we actually ship: review-sidecar, living-docs-auditor, compliance-verifier.
.claude/agents/* "overridden by us" (§6.2)	REWORK	post-C-1 we occupy zero AIF slots; added comm -12 disjointness probe.
/aif-verify should invoke best-practices-sidecar (§6.3)	REWORK	invokes AIF rules-sidecar (reads RULES.md) + aif-review (our skill-context).
docs-auditor (ours) (§3.4, item 13, item 16)	RENAME	living-docs-auditor (1:1 map).
best-practices-sidecar.react.md "inlined" stale-path probe (§2.4, item 6)	REMOVE	target never existed in current repo (T-AUD-B); §2.4 now checks the real post-C-1 invariant (no old agent file / no active wiring).
factory/RULES*.md (§2.1, §3.2, item 7, item 10, item 21)	REWORK	packages/preset-next-15-canonical/RULES*.md
scripts/audit-ai-docs.sh as source path (§1.5, §2.2, item 8, item 21, item 23)	REWORK	packages/core/audit-self/audit-ai-docs.sh. (Kept as CONSUMER path in §5 / Phase-5 install tests — T16: same string, different intent.)
tests/audit/audit-ai-docs.test.sh (§8.2, item 22)	REWORK	packages/core/audit-self/audit-ai-docs.test.sh; "16 pass" → "9 pass".
factory/{DESCRIPTION,ARCHITECTURE}, templates/shared/tsconfig.json (§3.1, §3.5, item 15)	REWORK	packages/core/templates/shared/
templates/react-next/ (§2.5)	REWORK	packages/preset-next-15-canonical/templates/
templates/ts-server/ (§2.5, §3.5)	KEEP	still at repo root — deliberately kept per Phase-3 Gate 2 (option A), templates/shared+react-next deleted.
§8.1 "no CI for the package" gap	REWORK	RESOLVED — .github/workflows/audit-self.yml exists (14 jobs).
§8.2 "no negative tests" gap	REWORK	RESOLVED for ts-server (9 pass/0 fail); react-next still uncovered → kept as finding.

Run evidence (T2/T3 — every reworked/kept mechanical probe executed against current main)

1.1 git ls-files | wc -l                → 390 ✓
1.5 executables                         → FAIL: setup.sh not +x  (REAL FINDING, see below)
2.1 orphan-rule check (R1-R11)          → no ORPHAN ✓
2.2 R/D comment count                   → 5 (≥5) ✓
2.4 stale agent/path                    → empty ✓ (no file, no wiring)
2.5 templates ls                        → ts-server (root) + preset templates present ✓
3.5 node --check eslint + tsconfig json → clean ✓
8.1 CI jobs grep                        → mechanical/rule-to-probe/probe-tests +11 more ✓
8.2 / item22 negative-test run          → 9 pass / 0 fail ✓
item9/item21 rule-to-probe replication  → exit=0, no ORPHAN ✓
item23 self-audit run                   → exit 0; 4 PASS, 0 FAIL, 2 WARN (no AGENTS.md / no tool-decisions.md — acceptable) ✓

Sanity: markdownlint 0 errors (fixed 2 pre-existing MD040 bare-fence violations in the copy-paste prompt block → text); both files ≤500 (453 / 302); 21 embedded bash blocks, 0 bash -n errors. Pre-push: principles 81 pass, render-drift clean, self-tests 16 pass.

Genuine findings surfaced (NOT fixed here — audit-tooling refresh only, per kickoff §7)

1. MINOR — setup.sh lacks the executable bit (git mode 100644; install.sh is 100755). Works via bash setup.sh, but inconsistent. §1.5 correctly FAILs on it (annotated as a known finding).
2. MAJOR — no negative-test file for the react-next probes (packages/preset-next-15-canonical/audit-self/audit-ai-docs.react-next.sh: R17 + the R12/R14/R15/R16a/R16b/R20 ESLint-delegated probes). The ts-server probes are negative-tested; the react-next ones are not. §8.2 / item 22 flag it.

§1.7

§1.7 Skipped: internal audit-tooling doc refresh consequent to an already-merged decision (C-1); no new rule, principle, or discipline is introduced — and discipline-self-check does not fire on root *.md (path filter).

AI-laziness traps honored

• T2/T3 — ran every reworked probe, pasted command output above; no prose-only "would pass".
• T16 / T-AUD-A — did not sed-swap: scripts/audit-ai-docs.sh kept as the consumer path in install-test sections vs re-pointed to packages/core/audit-self/ in source-audit sections; templates/ts-server kept (Gate 2) while templates/shared/react-next moved. Each reference mapped by intent.
• T-AUD-B — removed the dead best-practices-sidecar.react.md probe rather than inventing a new target.
• T14 — the green is real (executed), not assumed.
• T15 — these docs are the project's recursive self-audit; the refreshed audit run on current main is internally consistent (2 acceptable WARNs, 1 real finding correctly surfaced).