Decommission Checklist — Make a Shutdown Actually Stick

The problem

Killing a system or service is easy, but making that shutdown stick in an agent's instruction layer is hard. Agents often resurrect dead workflows because stale skill files, job descriptions, or loose drafts in the project root act as live instructions that override memory notes.

What it does

• Executes a top-to-bottom "kill order" that inserts precedence clauses into always-loaded instruction files like .cursorrules or CLAUDE.md.
• Performs comprehensive grep searches across skill files, job prompts, and descriptions to identify every hidden reference to the retired system.
• Tombstones retired skill descriptions to prevent silent re-creation and redirects agents to live replacements.
• Identifies and quarantines loose draft files or versioned skill fragments that act as accidental live instructions.
• Produces a mandatory PASS/FAIL verification report ensuring no stale references remain in the execution context.

Why this beats prompting it yourself

A simple "don't use X" prompt usually fails when a specialized skill file explicitly tells the agent to use X. This skill implements a structural hierarchy where the kill order outvotes cached instructions and physically modifies or archives the files that cause re-injection, ensuring the decommission is permanent.

Use cases

• Retiring a legacy API integration or third-party service safely.
• Shutting down a scheduled agent job without it being silently re-created.
• Cleaning up experimental skill drafts that are interfering with production workflows.
• Migrating from an old tool to a new one while ensuring the old path is never taken.

Known limitations

Cannot programmatically delete jobs or files in read-only environments; in these cases, it generates a manual checklist for the user.