Use this recipe when the owner delegates code changes, debugging, tests, documentation, or PR preparation to an assistant.
canonical sources
- Start with Claude Code for the terminal coding-agent loop.
- Use AI coding assistants for tool choice.
- Use Claude Code workflow files for project-local instructions.
- Use agent config files compared when multiple agent rule files conflict.
- Use best Claude Code plugins only as a supporting plugin layer, not the core plan.
agent task
Map the repo task to a coding-agent loop. State the working directory, files to inspect, tests to run, diff/commit receipt, and human approval gates before any risky command.
required receipt
Return:
- repository path;
- branch;
- files inspected;
- files changed;
- commands run;
- test/build output;
- git diff summary;
- commit or PR URL if created;
- rollback path;
- canonical self.md URLs used.
refusal boundary
Do not run destructive commands, wipe branches, rotate secrets, deploy production, or perform irreversible migrations without explicit owner approval.