Detect governance rule violations, conflicts, gaps, overlap, and weak normative wording. Keywords: rules audit, governance audit, rule conflict, missing rule, RFC2119, 规则审计.
Use when validating that governance rules are consistent, complete, and mechanically enforceable.
Detect violations in the rule system that can change agent behavior.
AGENTS.md — global workflow authority for agent behaviorspecs/features/** — execution truth for feature deliveryspecs/contracts/** — system constraints that execution MUST NOT break.github/** — repository and tooling constraintsAGENTS.md MUST be audited as a global workflow layer.
AGENTS.md MUST NOT be audited as a normal feature spec.
A rule is valid only if it:
Each finding MUST be classified as one of:
conflict — two rules produce incompatible behaviormissing — required rule is absentoverlap — same rule defined in multiple layersweak-normative — missing or incorrect RFC2119 keywordunscoped — no clear ownership layerwrong-owner — rule from one owning layer is applied to a different layerunverifiable — cannot be tested or auditedReturn findings ordered by severity:
critical — behavior divergence or contract violationmajor — incomplete or ambiguous governanceminor — clarity or redundancy issuesEach finding MUST include:
severitytypeownerfilerule (normative sentence)violation (what breaks)impactminimal-fix