living-docs-auditor
Runs scripts/audit-ai-docs.sh and reports findings. Catches backward Living-Documentation drift — whether AGENTS.md/RULES.md rules still hold in the actual code. Reports; does not fix.
living-docs-auditor
Status: shipped-beta · Ships to: npm lane — delivered into the consumer's .claude/agents/ by the setup delivery loop (setup.d/20-agents.sh:24) · Fires at: dispatched as a Claude Code sub-agent (frontmatter name:; the session's agent list carries its one-line description)
What it is
"> Authoritative for: living-docs-auditor sub-agent prompt — runs audit-ai-docs.sh and reports backward Living-Documentation drift …" between AGENTS.md/RULES.md rules and code; reporting-only (agents/living-docs-auditor.md:10). The job in one line: "You enforce code-vs-docs consistency: rules declared in AGENTS.md must hold in the actual code. …" (agents/living-docs-auditor.md:13). The name is the de-collision record: "# - docs-auditor — RENAMED to living-docs-auditor (de-collides with AIF's same-named agent)." (setup.d/20-agents.sh:14); the agent header carries the same note — "(Renamed from docs-auditor to coexist with AI Factory's same-named, different-job agent.)" (line 10 tail).
How it works
- Mechanism: "The mechanism is
scripts/audit-ai-docs.sh(orscripts/audit-ai-docs.react-next.shfor UI projects). …" (agents/living-docs-auditor.md:15). - "### Step 1: Detect which audit script applies" (line 23) — Next.js/React detection selects the react-next script (lines 27-33).
- "### Step 2: Verify the script exists" (line 37):
[ -f "$SCRIPT" ] || {(agents/living-docs-auditor.md:39) — graceful degradation, "echo "INFO: $SCRIPT not present in this project."" (line 41). - "### Step 4: Parse output" (line 53): "-
PASS: R<N>: <rule name>— rule satisfied (e.g.PASS: R7: Time/randomness injected via Clock/Random)" (line 57), "-FAIL: R<N>: <rule name>followed by indented violation lines (file:line: details)" (line 58); "-R1–R9map to rules in.ai-factory/RULES.md." (line 63). - Auxiliary drift checks (§5.1-5.5, line 128 heading) run even when the script is absent (line 170).
- "## Rules of engagement" (line 167): "- You don't modify code. You don't even modify the audit script. Only run and report." (line 169); "- WARN ≠ block. Decay-watch warnings inform but don't block. Only FAIL blocks
/aif-verify." (line 173).
Satellites & companions
- USES:
scripts/audit-ai-docs.sh/audit-ai-docs.react-next.sh(line 15) and.ai-factory/RULES.md/RULES.react-next.mdprobe mappings (lines 63-64). - ADAPTS: the AIF installer's script population in consumer projects — in the source project the script is absent and Step-2 degradation applies (frontmatter description block, lines 4-5).
- ADDS: the translation layer from raw probe output to PASS/FAIL/WARN verdicts plus the auxiliary awk/grep drift checks (lines 128-163); the reporting-only engagement rules (lines 167-174).
- Census family satellites: F28 (audit script family — the
audit-ai-docs.shscripts this agent executes).
Anchors
agents/living-docs-auditor.md:2—name: living-docs-auditoragents/living-docs-auditor.md:3—description: >-agents/living-docs-auditor.md:4—Runs scripts/audit-ai-docs.sh and reports findings. Catches backward Living-Documentation drift — whether AGENTS.md/RULES.md rules still hold …agents/living-docs-auditor.md:10—> **Authoritative for:**living-docs-auditorsub-agent prompt — runsaudit-ai-docs.shand reports backward Living-Documentation drift …agents/living-docs-auditor.md:15—The mechanism isscripts/audit-ai-docs.sh(orscripts/audit-ai-docs.react-next.shfor UI projects). …agents/living-docs-auditor.md:23—### Step 1: Detect which audit script appliesagents/living-docs-auditor.md:39—[ -f "$SCRIPT" ] || {agents/living-docs-auditor.md:53—### Step 4: Parse outputagents/living-docs-auditor.md:63—-R1–R9map to rules in.ai-factory/RULES.md.agents/living-docs-auditor.md:169—- **You don't modify code.** You don't even modify the audit script. Only run and report.setup.d/20-agents.sh:14—# - docs-auditor — RENAMED to living-docs-auditor (de-collides with AIF's same-named agent).setup.d/20-agents.sh:24—for f in "$PKG_ROOT"/agents/*.md; do
All anchors at framework pin b069c59328aa8e08671d3a19fd25efa618dfc1c5 (fetched 2026-09-11 from artyhoo/getff staging; ancestry of census pin a1337cb301 verified).
fidelity-auditor
Cold WHAT-conformance acceptance audit at a stage-PR boundary. Given ONLY the kickoff/spec (or a scoped section of it) and the 3-dot diff — NEVER the chat, the design dialogue, or the implementation log — judges whether the diff is what the kickoff asked for.
review-sidecar
Reviews diff as an external reviewer with no memory of how the code was written. Catches tautological tests, mock-only assertions, missing edge cases, React/Next anti-patterns. Reports; does not fix.