What problem does it solve? Documentation drifts from code over time, and manually reconciling docs against implementation across a large repository is slow and error-prone. This Skill runs a topic-scoped, report-only audit that reconciles EnvGene docs against code, schemas, and other docs, emitting a fact-oriented verdict table without editing any documentation. ## Core Features & Use Cases - Five-phase audit pipeline: scope resolution, claim extraction, truth location, verdict assignment, and report generation, with a user-confirmed corpus before auditing begins. - Fact-oriented verdict model: classifies each claim as contradiction, doc-ahead, doc-vs-doc (code breaks tie or code silent), unverifiable, or consistent, with a hard evidence-pair gate requiring doc and code file:line citations. - Parallel execution with re-verification: fans out Explore agents and per-slice workers for wide corpora, then re-verifies every delegated finding in the main context before it enters the report. - Use Case: Ask how EnvGene handles SSL certificates; the Skill assembles the relevant docs, the handle_certs.sh entry point, and schemas, extracts checkable claims, and returns a verdict table showing where docs and code diverge. ## Quick Start Run /discrepancy-audit with a topic such as "how EnvGene handles SSL certificates" to receive a verdict table reconciling the docs against the code.
