claims-conformance-auditor
Cold conformance audit of docs-site claims against shipped reality. Given ONLY the doc surface(s) to audit (never the authoring narrative), enumerates every factual claim the docs make about the repo/state, verifies each against the live source with command+output or file:line evidence.
claims-conformance-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: the claims-conformance audit protocol — claim taxonomy, inputs," per-claim verification method, output grammar (agents/claims-conformance-auditor.md:21-22). Class B cold detection layer: "> Class: B — the named cold-agent DETECTION layer for docs-claims drift. A release" checklist is merge authority, never detection (agents/claims-conformance-auditor.md:10-11). Fires "before merging or publishing any change that touches docs which assert facts" about the repository (agents/claims-conformance-auditor.md:15); in-framework it backs "> assembly gate (row F5)." of the beta-program spec §8 (agents/claims-conformance-auditor.md:19-20).
How it works
- Cold seat: "You are a COLD conformance auditor. You are dispatched with ONLY the doc surface(s) to audit" and paths only (
agents/claims-conformance-auditor.md:31); one question — does every factual claim match the repo right now (lines 33-34). - "## Claim taxonomy (what is in scope)" (line 45): "- counts — «10 agents», «all 20 hooks», «6 dirs at the default depth»;" (line 50), existence/shipped, behaviour, and path/process claims; normative prose and already-gated generated sections are out (line 56).
- "## Method (no prose-only findings — per the AI-laziness traps cited below)" (line 60): "1. Enumerate the claim population BEFORE verdicting (T10): read the audited surface" (line 62), then "2. Verify each claim against the LIVE source, never against another doc (T3 +" (line 66) — command output first,
file:linesecond; a second doc proves nothing (lines 69-73). - "3. Verdict per claim:" (line 74) — VERIFIED / GAP (quote both doc line and probe output) / UNVERIFIABLE-needs-human.
- "## Output format" (line 84): per-claim rows (line 95: " - file:line «<claim, verbatim, ≤15 words>» — VERIFIED (<file:line | command → output>)"), coverage line, "Overall: GO | REVISE | STOP — " (line 99); "
GAP→ REVISE; population too small, surface unreadable, or claims not enumerable → STOP;" (line 88).
Satellites & companions
- USES: reviewer-discipline clauses (line 38 heading) and the live-source posture from destination-environment-verification §1b (lines 66-68).
- ADAPTS: the dispatch-input-checker GO/REVISE/STOP verdict grammar (line 86-87).
- ADDS: the claim taxonomy + per-claim live-probe evidence rule; the dispatching session "folds
GAProws into fixes (owner-gated files get" patch proposals and cites the VERIFIED evidence (agents/claims-conformance-auditor.md:102-103). - Census family satellites: E6 (the encyclopedia's own claims-conformance pass dispatches this agent at a fresh pin); in-framework, the beta-program spec §8 assembly gate row F5 (lines 19-20).
Anchors
agents/claims-conformance-auditor.md:2—name: claims-conformance-auditoragents/claims-conformance-auditor.md:3—description: Cold conformance audit of docs-site claims against shipped reality. Given ONLY the doc surface(s) to audit (never the authoring narrative), …agents/claims-conformance-auditor.md:21—> **Authoritative for:** the claims-conformance audit protocol — claim taxonomy, inputs,agents/claims-conformance-auditor.md:31—You are a COLD conformance auditor. You are dispatched with ONLY the doc surface(s) to auditagents/claims-conformance-auditor.md:45—## Claim taxonomy (what is in scope)agents/claims-conformance-auditor.md:62—1. **Enumerate the claim population BEFORE verdicting** (T10): read the audited surfaceagents/claims-conformance-auditor.md:66—2. **Verify each claim against the LIVE source, never against another doc** (T3 +agents/claims-conformance-auditor.md:74—3. **Verdict per claim:**agents/claims-conformance-auditor.md:99—Overall: GO | REVISE | STOP — <one-line basis>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).
capability-reuse-auditor
Audits a proposed or just-authored new capability (a SKILL.md, an agent, or a packages/core module) for overlap with an existing own-stack or upstream capability, and checks that its Prior-art trailer's verdict matches what the body actually does. Flags reinvention before handoff. Reports; does not fix.
compliance-verifier
Reviews PR description §1.7 Forward-check and Backward-check sections for substantive evidence — file:line citations, sweep completeness, exemption quality. Reports; does not fix.