Uses agentic search to understand your entire codebase without manual context selection
Test, build, feedback so that the AI can check itself and iterate on its own
Good to help understand and get to know a code base
Good for adding individual features, one at a time, across multiple files
Optionally can let it make changes to multiple files, or can verify each change it makes
Can call your systems tools like bash and git and github so there is some risk. Ideally you should run it in a VM or in CI/CD to protect your system
Better with typed languages in my experience.
Install
curl -fsSL https://claude.ai/install.sh | bash
brew install claude-code
Commands & Usage
claude
Initiate a repo: /init to initiate the repo and have Claude create a claude.md file for the repo /agents
Ask to plan first:
Best workflow is to think in terms of features with end goals that can be verified
claude --permission-mode plan
Reference files to add to context manually - Explain the logic in @src/utils/auth.js
Pipe: cat build-error.txt | claude -p 'concisely explain the root cause of this build error' > output.txt
Project specific commands: mkdir -p .claude/commands
Claude.md
Have Claude help you write a claude.md file for your project