Playbook

1. Define the Trigger Boundary

Write one sentence that distinguishes tasks requiring the skill from adjacent tasks that do not. Use the terms an operator will actually use.

2. Specify the Output Contract

Name the artifact, required evidence, and stop condition. Replace “review the code” with a structured finding format and a definition of materiality.

3. Keep the Entrypoint Lean

Put the invariant workflow in the main skill file. Move domain references, long examples, and templates into linked resources. Put repeatable transformations in scripts.

4. Isolate Vendor Controls

Maintain one portable body where possible. Add thin product-specific metadata for discovery paths, invocation controls, tool permissions, model selection, and arguments.

5. Test Four Cases

• A clear positive trigger activates the skill.
• An adjacent negative case does not.
• Manual invocation passes inputs correctly.
• Context compaction or resume does not silently remove a load-bearing step.

Run these cases in every supported surface. “Uses AgentAgentA worker you delegate to: brief it, and it takes steps on its own. A chatbot answers; an agent acts.An LLM that runs tools in a loop toward a goal — it acts, checks the result, decides the next step, and repeats until done.Full definitionClose Skills” is not proof of identical runtime behavior.

6. Verify Side Effects

Scripts validate their inputs, quote paths, avoid ambient secrets, and produce reviewable output. Any external mutation requires the same authority it would require outside the skill.