Every CLI flag and keyboard shortcut organized by workflow.
claude # New session
claude --continue # Continue last session
claude --resume # Pick from recent sessions
claude --from-pr 123 # Start from a PR's context
claude --session-id <id> # Resume specific session
claude --fork-session <id> # Fork an existing session
claude --system-prompt "You are a Go expert"
claude --system-prompt-file ./prompts/go.md
claude --append-system-prompt "Also run tests after changes"
claude --agent researcher # Run with specific agent
claude --agents '{"name":"researcher","tools":["Read","Grep"]}'
claude --teammate-mode # Launch as a teammate in agent teamsclaude --permission-mode plan # Force plan mode
claude --allowedTools Read,Grep # Whitelist specific tools
claude --disallowedTools Bash # Blacklist tools
claude --print "explain this code" # One-shot, no interactive
claude --output-format json # JSON output for scripting
claude --json-schema '{"type":"object"}' # Structured output
claude --verbose # Debug outputclaude --max-budget-usd 5.00 # Cap spending
claude --max-turns 50 # Limit conversation turns
claude --worktree # or -w: auto-create git worktree
claude --add-dir ../lib # Add extra directory to context
claude --mcp-config ./mcp.json # Load MCP config
claude --strict-mcp-config # Fail if MCP server errors
claude --plugin-dir ./my-plugin # Load a plugin directory
claude --chrome # Enable Chrome integration
claude --no-chrome # Disable Chrome integration
claude --ide vscode # Set IDE context
| Key |
Action |
Enter |
Submit prompt |
Shift+Tab |
Cycle modes (Normal > Auto > Plan) |
Ctrl+C |
Cancel current generation |
Ctrl+L |
Clear screen |
Ctrl+B |
Send current task to background |
Ctrl+F |
Kill all background agents (two-press) |
Up/Down |
Navigate prompt history |
Esc Esc |
Rewind to last checkpoint |
| Key |
Action |
Shift+Down |
Navigate between teammates (wraps) |
Shift+Tab |
Toggle delegate mode |
| Command |
Purpose |
/clear |
Reset conversation |
/compact |
Compress context (do at task boundaries) |
/context |
Show context window usage |
/resume |
Resume a previous session |
/rename |
Rename current session |
/cost |
Show session cost (API key users) |
/usage |
Check plan limits |
/extra-usage |
Pay-as-you-go overflow billing |
| Command |
Purpose |
/model |
Switch model or effort level |
/fast |
Toggle fast mode |
/voice |
Toggle voice mode (hold spacebar to talk) |
| Command |
Purpose |
/permissions |
Manage tool permissions |
/agents |
Create and manage subagents |
/memory |
View/edit auto memory |
/doctor |
Diagnose configuration issues |
| Command |
Purpose |
/config |
Open settings |
/terminal-setup |
Configure terminal |
/statusline |
Customize status bar |
/keybindings |
Customize keyboard shortcuts |
/sandbox |
Configure sandboxing |
/vim |
Toggle vim keybindings |
| Command |
Purpose |
/export |
Export conversation |
/debug |
Toggle debug mode |
/rewind |
Rewind to checkpoint |
echo "explain this error" | claude --print
SESSION_ID=$(claude --output-format json --print "fix the bug" | jq -r '.session_id')
claude --session-id "$SESSION_ID" --print "now add tests"
claude --max-budget-usd 2.00 --max-turns 20 --print "review this PR"
ultrathink in your prompt triggers maximum reasoning effort/doctor diagnoses config issues — run when things feel off- Manual
/compact at 50% prevents the "agent dumb zone" near capacity
/rename your sessions — easier to find with /resume laterclaude -w is faster than manually creating worktrees- Voice mode (
/voice) — hold spacebar, speak, release to transcribe