Claude Code is an agentic coding tool that runs in your terminal (and can also integrate with IDEs). It can read your repo, propose edits, run commands/tests, and help with git workflows using natural language.
Use /init to generate or set up a CLAUDE.md guide for the project.
- “Explain the folder structure and where the entry point is.”
- “Add tests for X and run them.”
- “Refactor this module without changing behavior.”
Claude Code is permission-based and typically read-only by default until you approve edits/commands.
Use CLAUDE.md to store:
- Project rules (coding style, architecture constraints, “don’t touch these folders”)
- Preferred workflows (test commands, lint rules)
- Recurring context (API base URLs, feature flags)
Claude Code also supports memory across sessions. Keeping instructions organized (headings + bullet points) and updated helps Claude stay consistent.
/memory— edit the relevant memory /CLAUDE.mdfiles
- Claude Code can inspect your repo as needed during a session (you don’t always have to manually paste files).
- In IDE workflows (e.g., VS Code integration), you can mention files/line ranges and review plans before accepting edits.
Claude Code is designed to be transparent and permission-controlled:
- Read-only by default
- Requests explicit approval when it needs to edit files or run commands
/permissions— view/update permissions/sandbox— enable a sandboxed bash tool with filesystem/network isolation (fewer prompts, safer boundaries)/security-review— security review of pending changes on the current branch
Claude Code supports model selection, including an opusplan alias that uses:
- Opus for planning/architecture decisions
- Sonnet for execution/code generation
/model— select/change model
Claude Code has interactive slash commands for common actions.
/help— list available commands
It also supports custom slash commands as Markdown files (project-specific or personal), including namespacing by folders.
Claude Code can connect to external tools and data sources via Model Context Protocol (MCP). MCP servers can give Claude access to your tools, databases, and APIs.
/mcp— manage MCP server connections and OAuth authentication
Subagents are specialized assistants you can invoke for particular tasks, with:
- Custom prompts
- Tool configuration
- Separate context windows (helps keep focus and reduce clutter)
Claude Code includes workflows aimed at PRs and reviews, including GitHub app setup and viewing PR comments.
/help— usage help and command discovery/init— initialize project with aCLAUDE.mdguide/exit— exit the REPL/resume [session]— resume a conversation by ID/name/rename <name>— rename current session/export [filename]— export conversation to file/clipboard
/memory— edit memory /CLAUDE.mdfiles/todos— list current TODO items
/review— request a code review/rewind— rewind the conversation and/or code (checkpointing)/security-review— security review of pending changes
/permissions— view/update permissions/sandbox— enable sandboxed bash tool/doctor— checks installation health/status— open settings/status interface/privacy-settings— view/update privacy settings
/model— select/change model/output-style [style]— set output style
/mcp— manage MCP connections/ide— manage IDE integrations and status/install-github-app— set up Claude GitHub Actions for a repo/pr-comments— view pull request comments
/stats— usage and session history visualization/release-notes— view release notes/vim— vim mode
- What this repo does in 2–4 bullets
- Install:
- Dev server:
- Tests:
- Lint/format:
- Style rules
- Architectural constraints
- What not to change
- How to add a feature
- How to add a test
- How to cut a release
- If Claude feels slow or you’re working in a huge repo, use context-reduction workflows (for example, compacting context).
- Use
/helpearly and often to discover what’s available in your version.