capability-reuse-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: capability-reuse-auditor sub-agent prompt — the semantic overlap" triage of a proposed/just-authored capability plus the trailer↔body verdict-consistency check; reporting-only (agents/capability-reuse-auditor.md:11-13). The gap it fills: F1 can confirm a Prior-art: trailer exists but "carries a Prior-art: trailer, but it cannot tell whether the body actually reuses the" cited art or merely re-describes it (agents/capability-reuse-auditor.md:30). Motivating incident: "Origin: PR #858 shipped .claude/skills/night-mode/SKILL.md re-describing the executor + dual-" reviewer loop with a correct-sounding trailer that passed F1 (line 33).

How it works

• "## Input" (line 43) — either a proposed capability or a just-authored file: "just-authored** capability file (path to a new SKILL.md, agents/*.md, or" (agents/capability-reuse-auditor.md:46).
• "## Method (no prose-only findings — per T3)" (line 50): name the problem-class functionally (step 1), enumerate the overlap set with counts (step 2, line 54: "2. Enumerate the candidate overlap set (state counts before judging — T10):") over the SSOT (line 55: "- Grep/Glob the SSOT (docs/meta-factory/prior-art-evaluations.md)") and .claude/skills/ + agents/ (line 59), then "3. For each overlap candidate, apply the problem-class match test (T16): «Upstream/existing" (line 63).
• Trailer check (step 4, line 66: "4. Trailer↔body consistency (if a trailer/verdict is supplied): does the body do what the") — reuse verdicts must "subordinate" to the owner, not claim authority over it (lines 67-70).
• Coverage honesty: "5. Distinguish «no overlap» from «low coverage» (T14): partial coverage reported as partial," (line 71).
• "## Verdicts you recommend (with GO/REVISE/STOP overlay per dispatch-input-checker.md §Output grammar)" (line 74) — REUSE-EXISTING / THIN-ADAPT / BUILD-JUSTIFIED / TRAILER-BODY-MISMATCH / INCONCLUSIVE, output grammar at line 93: "VERDICT: <REUSE-EXISTING | THIN-ADAPT | BUILD-JUSTIFIED | TRAILER-BODY-MISMATCH | INCONCLUSIVE>".

Satellites & companions

• USES: the prior-art SSOT (docs/meta-factory/prior-art-evaluations.md, line 55) and the reviewer-discipline clauses it imports (line 37: "## Reviewer-discipline clauses (reviewer-discipline.md §1+§2)").
• ADAPTS: the dispatch-input-checker GO/REVISE/STOP verdict overlay (line 74; also line 22: "> overlay on verdicts per dispatch-input-checker.md §Output grammar.").
• ADDS: the reuse-vs-reinvent judgment the deterministic F1 gate cannot make (line 30) and the verdict table — e.g. "| REUSE-EXISTING | An own-stack/upstream capability already covers this problem-class; …" (line 79) and the mismatch row (line 82).
• Census family satellites: C2 (reviewer-discipline — its §1+§2 clauses are imported verbatim at line 37).

Anchors

• agents/capability-reuse-auditor.md:2 — name: capability-reuse-auditor
• agents/capability-reuse-auditor.md:3 — description: Audits a proposed or just-authored new capability (a SKILL.md, an agent, or a packages/core module) for overlap …
• agents/capability-reuse-auditor.md:11 — > **Authoritative for:** capability-reuse-auditor sub-agent prompt — the semantic overlap
• agents/capability-reuse-auditor.md:37 — ## Reviewer-discipline clauses (reviewer-discipline.md §1+§2)
• agents/capability-reuse-auditor.md:43 — ## Input
• agents/capability-reuse-auditor.md:50 — ## Method (no prose-only findings — per T3)
• agents/capability-reuse-auditor.md:63 — 3. **For each overlap candidate, apply the problem-class match test** (T16): «Upstream/existing
• agents/capability-reuse-auditor.md:66 — 4. **Trailer↔body consistency** (if a trailer/verdict is supplied): does the body _do_ what the
• agents/capability-reuse-auditor.md:74 — ## Verdicts you recommend (with GO/REVISE/STOP overlay per dispatch-input-checker.md §Output grammar)
• agents/capability-reuse-auditor.md:93 — VERDICT: <REUSE-EXISTING | THIN-ADAPT | BUILD-JUSTIFIED | TRAILER-BODY-MISMATCH | INCONCLUSIVE>
• 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).