Team Starter Kit (Minimal Working Setup)
The hardest part of rolling out Claude Code in a team is rarely “missing features”—it’s everyone having a different workflow: some only describe verbally, some start editing immediately, some never run tests. The Starter Kit is a minimal setup that standardizes three essentials:
- clarify the problem first (Plan)
- make the change correctly (TDD/Build Fix)
- manage risk before merge (Code Review/Security)
This structure is designed for a practical team rollout: minimal files, clear boundaries, and easy iteration.
Directory layout (commit to repo; shared across the team)
your-repo/
├─ CLAUDE.md
└─ .claude/
├─ settings.json
├─ rules/
│ ├─ security.md
│ ├─ testing.md
│ └─ coding-style.md
├─ agents/
│ ├─ planner.md
│ ├─ code-reviewer.md
│ └─ build-error-resolver.md
└─ commands/
├─ plan.md
├─ tdd.md
├─ build-fix.md
└─ code-review.md1) CLAUDE.md: write down project facts
This file is “project memory”. Keep it executable and verifiable:
# Project overview
- Tech stack: …
- Entrypoints / structure: …
## Common commands (must be accurate)
- Install deps: …
- Run tests: …
- Build: …
- Lint/format/typecheck: …
## Constraints (team agreement)
- Default: analyze in Plan Mode before editing
- Any behavior change must add tests and/or docs
- Credentials/permissions changes require security review before merge2) rules/: split bottom lines into small files
The community practice: break rules into reviewable chunks. Start with three:
rules/security.md: secrets, input validation, least privilegerules/testing.md: TDD cadence, required coverage for critical pathsrules/coding-style.md: file size, function size, error handling, etc.
Don’t paste “a huge idealized policy”. Write the checklist your team will actually follow.
3) agents/: delegate “brain-switching” tasks
Three common starter agents:
planner: phased plan with risks/deps/acceptance criteriabuild-error-resolver: fix one build error at a time to avoid chain reactionscode-reviewer: severity-graded security and quality review over uncommitted changes
Don’t build too many agents initially—locking in planning/troubleshooting/review yields immediate gains.
4) commands/: workflows as /plan, /tdd, /build-fix, /code-review
Command files are workflow templates. The key is enforcing sequence.
Next steps
- Operationalize the loop: see Quality gates
- Turn team practices into reusable Skills: see Skill Pattern Library