What problem does it solve? Packaged composition skills (packaged/*/SKILL.md) can drift from what was actually measured during evaluation, and there is no guarantee their prose, steps, and quality bars still hold when executed. This Skill closes that gap by validating each packaged skill's structure cheaply and, when requested, executing it end-to-end exactly as it ships. ## Core Features & Use Cases - Structural validation: Runs node dist/e2e/cli.js --validate to check every packaged SKILL.md is well-formed, printing OK/FAIL per file with no agent spend. - Live dispatch E2E: Parses each packaged skill, writes per-case assertions files, and invokes the dispatch runner so one subagent executes each step on its declared _Model:_ model, honoring the tested == shipped == measured principle. - Scored verdicts: Scores each run against embedded assertions and the stamped quality bar, writes a per-case report.md, and reports an overall PASS/FAIL with failing cases. - Use Case: After editing a packaged composition skill, ask for a live E2E run to confirm the shipped artifact still passes its assertions and quality bar before releasing it. ## Quick Start Run the structural validation across all packaged skills and report which ones pass or fail.
