| name | haipipe-application-claim-audit |
| description | Claim audit for the intervention lifecycle. Verifies that every claim in the intervention's artifacts is traceable to an adopted A entry (A -> C -> anchor) and that no claim exceeds the evidence scope. Parallel to paper's claim-audit skill. Trigger: claim audit, verify claims, evidence check, /haipipe-application claim-audit. |
| argument-hint | [intervention-path] |
| allowed-tools | Bash, Read, Grep, Glob, Skill |
| metadata | {"version":"0.1.2","last_updated":"2026-07-17","summary":"Claim-evidence audit — every artifact claim traces to an adopted entry (A -> C -> anchor) and stays inside its evidence scope. History: ./CHANGELOG.md."} |
Skill: haipipe-application-claim-audit
Cross-references the intervention's artifacts against the claim
ledger and KB to catch:
- Orphan claims: artifact says something not in the claim ledger
- Overclaims: artifact states a claim more strongly than evidence supports
- Stale citations: artifact cites a K/W entry that has been superseded
- Missing citations: factual statement with no K/W backing
Audit scope
Reads:
0-lifecycle/1d-advice/1d-advice.md (design advice -- the A<-C chain)
0-lifecycle/1c-claims/1c-claims.md (claim ledger)
0-artifacts/*.md (drafted artifacts)
Writes:
0-artifacts/CLAIM_AUDIT.md (audit report)
Trace chain: every artifact move -> the A<n> it executes -> the C<n> each A derives from -> the claim's anchor. A break anywhere in artifact -> adopted A -> C -> anchor is a finding.
Audit checklist per artifact
For each factual statement in the artifact body:
[ ] Statement maps to a claim C## in the ledger
[ ] Claim C## status is supported or weak (not GAP)
[ ] If weak: artifact qualifies the statement appropriately
[ ] Cited K/W entry exists and is active (not superseded)
[ ] Statement does not exceed the K entry's scope
(e.g., K says "in high-variability patients" but artifact
generalizes to "all patients")
Risk profile
READ-ONLY on all source files. WRITES one audit report.
