Instructions
Use this skill when you need to introduce or scale Claude Code in a large codebase (monorepo or enterprise repo).
Start with navigability
- Prefer starting sessions from the relevant subdirectory instead of the repo root.
- Keep a lean root
CLAUDE.mdwith global conventions. - Add subdirectory
CLAUDE.mdfiles for local conventions.
Treat the harness as first-class
- Prioritize extension points in a layered order:
CLAUDE.md→ hooks → skills → plugins → MCP servers. - Add LSP integrations to improve symbol-level navigation (especially for C/C++ and multi-language repos).
- Use subagents to split exploration (mapping) from editing.
- Prioritize extension points in a layered order:
Use the right tool for the job
- Deterministic checks (format/lint/test) should be done with hooks/scripts, not with prose reminders.
- Put specialized workflows into skills so they don’t bloat every session.
Maintain configuration as models evolve
- Review
CLAUDE.md, hooks, and skills every 3–6 months. - Revisit anything created to compensate for older model limitations.
- Review
Define ownership & governance
- Assign a DRI (or an agent manager function) for Claude Code configuration, permissions policy, and plugin governance.
- Establish cross-functional alignment early (engineering + infosec + governance).
Examples
- You are rolling Claude Code out to a monorepo with many teams: use the checklist in examples/rollout-checklist.md.
- You are hitting navigation errors in a typed multi-language codebase: prioritize LSP integration and document it in your root
CLAUDE.md.