What problem does it solve? Auditing a growing collection of Agent Skills by hand is error-prone: dependencies between skills, unrouted mentions, missing routing evals, and oversized entry points are hard to see from raw Markdown. This Skill turns a skills repository into a deterministic, queryable inspection layer without changing the authored SKILL.md files. ## Core Features & Use Cases - Static projection: Discovers every SKILL.md, its resources, authored Markdown links, routing eval cases, and semantic sibling-skill mentions, then emits a disposable OKF bundle with RFC 0006 DuckDB schema contracts. - Graph and relational analysis: Compiles the bundle with okf-parser to produce dependency graphs and a DuckDB database, keeping authored links distinct from reviewed mentions with full source provenance. - Canonical audit views: Ships typed SQL views for eval coverage, resolved skill relations, mentions without edges, isolated skills, and resource surface, exported as a JSON report. - Use Case: Point it at a skills monorepo to find which skills lack routing evals, which are isolated from the dependency graph, and which scripts are never referenced, before deciding what to refactor. ## Quick Start Ask the agent to project this skills repository with scripts/project.py into a derived bundle, compile it with okf-parser into DuckDB, and run the typed audit to report eval coverage and isolated skills.
