Claude Code for builders, in one page.
Claude Code is Anthropic's coding agent: you brief it in plain language in your terminal, and it reads your repo, edits across files, runs commands and tests, and works step by step behind a permission dial you control. Think of it as a fast, tireless junior who needs a clear brief and a careful review. It multiplies your strong people on well-defined work; it doesn't stand in for their judgement.
It does its best work on bounded, checkable jobs you can describe and verify: a feature inside a design you've set, tidy refactors, adding tests, a bug with clear repro steps, codebase questions, multi-file changes, a one-off script. Keep design, security-critical paths and ambiguous product calls with your people.
- Goal: one clear outcome, written like a ticket for a capable junior.
- Context and files: the area to work in, plus how to build, test and lint.
- Done: the test that passes, the behaviour you expect.
- Out of scope: what to leave alone. Can't scope it tightly? Break it down or keep it yourself.
- Permissions are your dial: grant the least the task needs and keep a yes in front of risky actions.
- Never expose production secrets. Use scoped, least-privilege tokens for any access.
- Connect tools via MCP only as a task needs, least privilege, sources you trust.
- Read every diff or pull request like a contributor's. Green tests are a start, not a sign-off.
- Work on a branch; protect main with required status checks and a required human review.
- Keep a CLAUDE.md instructions file in the repo. It pays off on every task after.
- 1. ......................................................
- 2. ......................................................
- 3. ......................................................
Scope, run, review, merge. Tighten the brief or the context when something comes back off, and keep what works. Always a human on the diff.