feat(agents): resolve C-1 agent-name collision with AI Factory (KEEP-AIF / skill-context / rename)

[artyhoo]

Summary

Resolves the C-1 agent-name collision with AI Factory (AIF v2.11.0), which ships three sub-agents whose filenames collide with ours (best-practices-sidecar, docs-auditor, review-sidecar). Default install silently let AIF's thin stubs win; --force would strip AIF frontmatter its pipeline depends on. The accepted resolution (per research-patch) treats the three non-uniformly so our install occupies zero of AIF's agent slots — adopting AIF where it already covers the job, riding AIF's native skill-context override where our content is unique, and renaming where the job genuinely differs.

Changes

• best-practices-sidecar → KEEP-AIF (ADOPT): excluded from the install glob (install.sh:235-241); removed from extension.json + REQUIRED_HEADER_DOCS. AIF's own rules-sidecar already reads .ai-factory/RULES.md; our edit-time ESLint + pre-push are the earlier-channel enforcers. Its only residue with no earlier channel (R10 naming, R4/R17 test-existence) ships via a new aif-rules-check skill-context override. The file is kept in-repo as portable SSOT (not deleted).
• review-sidecar → PRESERVE via skill-context (ADAPT): anti-tautology content delivered into AIF's aif-review pipeline via packages/core/templates/shared/skill-context/aif-review/SKILL.md; agents/review-sidecar.md kept as portable SSOT with @dual-pair/spec-of markers (review-sidecar.md:12).
• docs-auditor → RENAME to living-docs-auditor (KEEP-NARROW): distinct job from AIF's docs-auditor (backward Living-Docs drift via audit-ai-docs.sh vs forward /aif-docs generation gating).
• New SSOT entry #50 — "skill-context = AIF-native vendored-agent override", verdict ADOPT.
• Doc-prose honesty sweep across README, INSTALL, INSTALL-FOR-AI, setup.sh, RULES*.md, skills/references/* + the two .claude/rules citation lists + CLAUDE.md ownership row, to match the new shipped set. Frozen/historical docs (docs/audits/**, closed-questions, open-questions, self-application, retros, research-patches) deliberately untouched.

Note: the load-bearing claim — does a background:true, maxTurns:6 sidecar actually read+apply skill-context? — was verified by a live headless probe before this PR (3/3 read, 2/2 apply with legitimately-phrased conventions; injection-shaped text correctly refused). So the skill-context path is confirmed, not assumed; the MERGE fallback is not needed.

Prior-art consult

• Capability commit carries a Prior-art: trailer citing prior-art-evaluations.md#50.
• New capability area surfaced → SSOT entry chore(deps): bump js-yaml and markdownlint-cli2 #50 added in the same commit with Verdict / Rationale / Trigger.
• No existing entries needed Last reviewed bump.
• Research used DeepWiki + live probe + CC docs (no context7 needed — this is a CC/AIF-internals adoption, evidence in the research-patch §2-§6).

Test plan

• npm run --prefix packages/core test:principles green (81 pass / 4 skip) — incl. principle 09 (REQUIRED_HEADER_DOCS ↔ install.sh SHIPPED_DOCS set-equality, length → 15) and principle 11 (capability/SSOT enforcement).
• Full core suite: 47/48 files pass; the 1 fail is validate-batch-spec soft-mode timing out on gh (environmental sandbox, gh-dependent — unrelated; CI has gh).
• npm run typecheck clean; npm run check:skill-drift PASS (0 broken refs, all frontmatter present).
• markdownlint-cli2 clean on all staged .md (pre-commit hook).

§1.7 Forward-check applied

Checked the resolution against every relevant existing discipline:

• build-first-reuse-default verdict typology (build-first-reuse-default.md:22, :25) — each agent got an explicit ADOPT / ADAPT / KEEP-NARROW verdict with a T16 problem-class match.
• dual-implementation-discipline §5/§7 (dual-implementation-discipline.md:115) — review-sidecar + best-practices-residue treated as one-SSOT-two-channels, bound by @dual-pair/spec-of markers (review-sidecar.md:12, best-practices-sidecar.md:12).
• doc-authority-hierarchy §2 — both new shipped skill-context templates carry Authoritative-for headers and were added to REQUIRED_HEADER_DOCS (09-doc-authority-hierarchy.ts:93).
• no-paid-llm-in-ci — the version-bump safety check is deterministic comm/grep (research-patch §5.3); the live probe ran on the operator's subscription, not paid API.

§1.7 Backward-check applied

Swept the complete existing surface, not examples:

• grep -rn for all three agent names across the repo → enumerated ~16 live files + the frozen set; updated every non-frozen reference and verified grep for stray docs-auditor (our agent) outside frozen returns empty.
• Principle 09's install.sh↔REQUIRED_HEADER_DOCS set-equality test re-run green after the length change (09-doc-authority-hierarchy.test.ts:164).
• Confirmed AIF's install ships agents via a glob, not the SHIPPED_DOCS array — so KEEP-AIF required a glob exclusion, not just an array edit (install.sh:240).

[artyhoo]

Superseded by #82. PR #79 (merged in parallel) already landed the overlap (KEEP-AIF best-practices + RENAME docs-auditor→living-docs-auditor + prose sweep); #80 conflicted by redoing it. #82 is the clean delta-only follow-up off current main — the skill-context delivery + SSOT #50 that #79 explicitly deferred pending the live probe (which has now run: 3/3 read, 2/2 apply). Best-practices file: accepted #79's deletion; residue preserved via aif-rules-check skill-context.