agent-runbook-skill
When to Use
Use this skill when an agent receives a human Markdown runbook and needs a dry-run action plan before touching files, tools, or external systems. It may read local runbook files only. It must not execute the runbook. External writes, deploys, publishes, messages, destructive commands, and approvals remain outside the skill boundary and require explicit user approval. Imperative destructive actions are classified as approval-required; command-shaped git push, npm publish, gh pr create, gh issue create, gh release create, and gh pr merge actions are classified as external-write, including after Run / Execute and supported sudo, doas, command, or assignment-bearing env prefixes. Command-shaped git commit, package test/build commands (npm, pnpm, yarn, and bun), the allowlisted npm run lint, npm run check, npm run smoke, npm run release:check, npm run release:readiness, and npm run package:smoke verification scripts, and cp, mv, touch, and mkdir actions are classified as local-change, including explicit Run / Execute wrappers and then / and then sequences. Other package scripts are not inferred to be local changes. Inspection and documentation wording that only discusses these operations remains read-only. Validate by running the fixture smoke command and reviewing the action classes before acting.
Required Inputs
Local files supplied by the user or repository fixtures.
Side Effects
Read-only. The CLI writes results to stdout only.
Validation
Run npm test, npm run check, and npm run smoke.