← ClaudeAtlas

evidence-auditorlisted

Power skill that confirms every claim tagged as verified in a lifecycle artifact cites a resolvable evidence source.
JeelVankhede/agentsmyth · ★ 0 · Code & Development · score 72
Install: claude install-skill JeelVankhede/agentsmyth
# Evidence Auditor ## Purpose For every claim in a lifecycle artifact that is tagged as verified, confirm a cited evidence source exists — command output, a file path, a review finding, or another artifact reference — and flag claims with no citation. Enforces `agent-behavior.yaml`'s `evidence_policy.no_external_claim_without_evidence`. This is a power skill, not a lifecycle phase. It is gate-bound: it runs at Review, Test, Ship, and Reflect Exit Gates. ## Invocation Context Use this skill when: - Review is finalizing findings that assert something passed, failed, or was checked - Test is recording verification outcomes - Ship is aggregating evidence for a go/hold/blocked recommendation - Reflect is describing what worked, what didn't, and outcomes Do not invoke it for claims explicitly marked as opinion, recommendation, or open question — only claims asserted as fact or verified state are in scope. ## What To Load **Foundation** (confirm in context; load if not already present): - Root `AGENTS.md` - `workflow/router.md` - `workflow/lifecycle.md` - `workflow/rules.md` **Minimum for invocation**: - This file - `references/output-schema.md` **Before starting work**: - `references/citation-shapes.md` — what counts as a resolvable citation - `workflow/agent-behavior.yaml` — `evidence_policy` block **Load when the step requires it**: - The artifact being audited, in full ## Inputs - The lifecycle artifact (or draft) containing claims to audit. - `agent-behavior.yaml`