What problem does it solve? Documents like specs and plans are executed by literal-following workers who get stuck on ambiguity, contradictions, and missing acceptance criteria. This Skill provides the shared audit standard — eleven failure modes, four evidence shapes, and severity calibration — so spec and plan audits apply one consistent bar instead of diverging copies. ## Core Features & Use Cases - Eleven sequential failure-mode passes: recommendation-coherence, internal contradiction, duplication, unsupported independence, argument soundness, constraint completeness, fix actionability, drift, scope creep, structural consistency, and metadata completeness. - Strict evidence grounding: every finding must use a doc quote, absence reference, wrong-claim, or internal-coherence shape, prefixed with its source heading or file path. - Immutable round recording: findings are registered via source_add against the audited document revision, and a JSON block is returned as the final response — never a written document. - Use Case: Dispatched as one round of sdlc-spec-audit, load this Skill first, read the stored spec revision with document_present, run the eleven passes, and register the round as a source the next spec revision must cite. ## Quick Start Load the sdlc-audit-criteria skill, then audit the stored revision of the given spec or plan through all eleven failure modes and return the findings JSON.
