Claude Code

Conclusion

Use Claude Code when you want an agent that lives in the repo, can plan, edit, and re-check against a command you define.

Prerequisites

• Official Claude Code install for your OS
• A project that can run with one clear verify command (tests, lint, or smoke script)

Steps

1. Install and authenticate per Anthropic’s current docs.
2. Add a short project rules file that states goals, forbidden paths, and the verify command.
3. Prefer plan mode for multi-file work; only implement after the plan matches your acceptance criteria.
4. After mistakes, write the correction into the rules file so the next session inherits it.

Template fields

• Verify command:
• Forbidden paths:
• Definition of done:

Definition of done

You can start a session, state a scoped task, and the agent can run your verify command without you pasting secrets.

Anti-patterns

• Pasting API keys or .env contents into the chat
• Asking for “just rewrite everything” without a spec or verify loop

Sources

See frontmatter publicSources.