Give the working agent the objective, constraints, available gates, and authority to choose its method. Derive completion from the user's request and repository contracts. CI, independent review, commits, or publication belong in completion only when those contracts require them.

Read references/procedure.md before authoring. Inspect the repository and preserve the provenance of its commands and requirements. Explain material gaps before treating the skill as usable. Do not impose a phase graph or create queue or ownership infrastructure.

The generated skill must require method assessment when meaningful feedback exposes a weakness, and at completion. The agent chooses a useful repair or justified retention within task authority and remaining budget. It applies repairs to the current work, verifies affected results, and preserves useful lessons. Explicit stops and required approvals continue to govern the work.

Write to , unless the user or repository names another contributor-skill location. Keep the skill usable with the project's own tools; it must not depend on this authoring skill after generation.

Finish when the file expresses the contract, required commands are verified appropriately, and unresolved gaps are reported. Report what was checked. If the user also requested a trial, continue that authorized trial and use its observations to revise the skill.