Claude Code cheat sheet

Generate or improve a project CLAUDE.md based on the current repo. Set CLAUDE_CODE_NEW_INIT=true for an interactive flow that also sets up skills, hooks, and personal memory.

💡 Run this early in a new repo, then trim aggressively so only high-value instructions remain.

Enter plan mode directly from the prompt for read-only analysis and planning.

💡 Use before migrations, refactors, or any task that spans multiple files or uncertain requirements.

Compress the conversation to recover context while preserving the important thread.

💡 Give compact instructions; otherwise Claude may preserve the wrong parts of the conversation.

Visualize current context usage as a colored grid. Shows optimization suggestions for context-heavy tools, memory bloat, and capacity warnings.

💡 Check this before long sessions get sloppy; it's an early warning system for degraded performance.

Ask a quick side question without adding it to the conversation history or affecting the main task.

💡 Use this whenever you have a quick clarifying question mid-session that doesn't belong in the thread.

View and manage hook configurations for tool events (PreToolUse, PostToolUse, SubagentStart, Stop, etc.).

💡 Hooks are deterministic guardrails. Wire up formatters, linters, or safety checks here instead of relying on prompts.

Make the current local session available for remote control from claude.ai or the Claude mobile app.

💡 Use this when you start a session locally and want to hand off to mobile, or pick it up from another device.

Manage permission rules and allowlists for commands/tools.

💡 Allowlist safe repeat commands like test/lint/git read operations to avoid approval fatigue.

Create, inspect, and manage custom subagents.

💡 Make a read-only reviewer or debugger first; they pay off quickly and keep your main context clean.

Manage Claude Code plugins: install, uninstall, list, and reload.

💡 Plugins can ship hooks, skills, MCP servers, and agents together as a unit.

Fetch and display comments from a GitHub pull request. Auto-detects the PR for the current branch, or pass a PR URL or number. Requires the `gh` CLI.

💡 Follow up immediately with 'Address all unresolved review comments' to action them in one flow.

Create a branch (fork) of the current conversation at this point. Alias: /fork.

💡 Use when you want to explore an alternative approach without losing your current thread.

Switch models during a session. For supported models, use left/right arrows to adjust effort level inline.

💡 Switch up only for genuinely hard reasoning tasks; keep default or Sonnet for routine coding.

Change adaptive reasoning depth. `low`/`medium`/`high` persist across sessions; `max` is current session only (Opus 4.6). `auto` resets to model default.

💡 Often a better move than switching models when the task just needs deeper thought.

Manage CLAUDE.md memory files and auto-memory behavior. View, edit, or disable auto-memory entries.

💡 If Claude repeats the same mistakes across sessions, fix your memory instead of re-explaining manually every time.

Open an interactive diff viewer for uncommitted changes and Claude's per-turn edits. Use arrows to browse files and turns.

💡 Use this before commit/PR creation so you catch broad or unintended edits early.

Show token usage statistics for the session.

💡 If cost spikes, it usually means context sprawl, over-broad exploration, or using a stronger model than the task needs.

Start fresh and clear conversation history. Aliases: /reset, /new.

💡 For a truly messy session, sometimes a clean restart beats trying to salvage bloated context.

Run diagnostics on Claude Code installation and settings.

💡 This is the first thing to try before spelunking through PATH, auth, or config issues by hand.