fix(install): consumer-install-hardening F9b — /aif-* honesty sweep (remaining docs)

[artyhoo]

consumer-install-hardening — F9b: /aif-* honesty sweep (remaining docs)

Follow-up to #480 (F9) — maintainer-invited. F9 reframed AGENTS.md + checks-map.md; F9b finishes the sweep across the rest of the shipped doc surface.

The lie (same defect class as F9)

Three more shipped docs presented /aif-verify as the UNCONDITIONAL "REQUIRED before commit" gate the installer never delivers (.claude/commands/ is never shipped):

• packages/core/templates/shared/CLAUDE.md.template:22/:26 (ships to consumer root CLAUDE.md) — /aif-verify — REQUIRED before commit + Don't bypass /aif-verify.
• packages/core/templates/shared/DESCRIPTION.template.md:61/:65 (ships to .ai-factory/) — identical.
• skills/rules-as-tests/references/ai-traps.md:227/:254 — teaches consumers to write the lie in their own AGENTS/CLAUDE (the sample-doc template).

Fix

Reframe to conditional "if you use AI Factory (aif)" (not bundled); the unconditional pre-commit/PR gate now names what install actually ships (./scripts/audit-ai-docs.sh + .husky/pre-push + CI). /aif-verify demoted to a convenience wrapper. The 4 other shipped docs that mention /aif-verify (SKILL.md, overview.md, self-testing-docs.md, doc-organization.md) do so descriptively ("AI Factory /aif-verify pre-PR gate") — honest, left untouched (scope discipline; T16 — mentions ≠ lies).

Probe — broadened (closes F9's 2-doc-scope caveat)

Extended tests/install-sh/f9-aif-commands-honesty.test.sh to sweep all 37 shipped markdown docs. The audit-self.yml F9 step already runs this file — no workflow change.

§1.7 Forward-check applied

Same reframe pattern as F9 #480; complies with companion-install-principle.md §1 (companions install via their own installer — we ship NO /aif-* commands), build-first-reuse-default.md §1 (REUSE the already-shipped audit-ai-docs.sh + pre-push gate; only the bash probe extension is new), and no-paid-llm-in-ci.md (the probe is pure bash, zero API). Evidence: packages/core/templates/shared/CLAUDE.md.template:17, skills/rules-as-tests/references/ai-traps.md:227, tests/install-sh/f9-aif-commands-honesty.test.sh:108.

§1.7 Backward-check applied

Extends #480 rather than superseding it; the 4 descriptive-mention docs are intentionally untouched (verified zero lie-pattern hits). No existing rule superseded. The broadened sweep closes the 2-doc-scope caveat the F9 probe documented at tests/install-sh/f9-aif-commands-honesty.test.sh:11; the benign-vs-lying separation was proven before edit (packages/core/templates/shared/DESCRIPTION.template.md:61 matches, skills/rules-as-tests/SKILL.md:124 does not).

Verify-trace

• bash tests/install-sh/f9-aif-commands-honesty.test.sh → PASS=10 FAIL=0; both sweep neg arms bite
• lie-patterns verified zero-hit on the 4 benign docs + 2-hit each on the 3 lying docs (separation proven before edit)
• bash -n clean; portable (no mapfile — runs on macOS bash 3.2 + ubuntu CI)
• branch 3-dot diff vs staging = 4 F9b paths only; F9/F8 files untouched
• CI green on this PR — pending re-run after this body fix

Scope-out

Branch based on live staging 4bff351a; pushed via harvest-via-api.sh (git transport tunnel-blocked). Part of the consumer-install-hardening umbrella; runs in parallel with the issue-#478 S3 track (no file overlap).