After the basics, you can customize and automate Claude Code to fit your workflow: output styles, status line, scheduled tasks, sandboxing, building plugins and much more.
| Feature | How | What it does |
|---|
| Output styles | /config (output style) | Default, Explanatory (explains things), Learning (makes you write the code), or a custom style |
| Status line | /statusline or a statusLine command in settings | Info like model, branch, cost, context at the bottom |
| Keybindings | ~/.claude/keybindings.json | Rebind shortcuts, chord bindings |
| Model aliases | /model or --model opus | opus, sonnet, haiku; opusplan (plan with Opus, do the work with Sonnet) |
| Fast mode | /fast | Opus with faster output |
| Effort | /model / settings | From low to max; xhigh is the default for coding |
| Theme / vim mode | /config | Terminal look, vim keys |
- Auto memory: Claude Code itself saves important facts (user preferences, feedback, project context) in the project memory directory and recalls them in the next session from the MEMORY.md index.
- Checkpoints / /rewind : a snapshot of the code before every prompt; if it goes in the wrong direction, restore the code and/or the conversation.
- /resume, claude -c : continue old sessions; session history is saved locally.
- /compact and auto-compaction: a summary when the context fills up.
| Feature | What it does |
|---|
| /loop | Run a prompt or command repeatedly on an interval ("check the deploy every 5 min") |
| /schedule (routines) | Agents that run on a cron in the cloud, even when your laptop is off |
| Background tasks | Long commands (server, build) run in the background; Claude is notified when they finish |
| Monitor | Wait for a condition/log without polling |
| Headless (-p) | CI/CD, git hooks, scripts: claude -p "..." --output-format json |
| GitHub Actions / GitLab CI | PR review, issue to PR, @claude mentions |
| Hooks | Deterministic automation (Section 16) |
| Feature | Detail |
|---|
| Built-in subagents | Explore (read-only fast code search), Plan (implementation plan), general-purpose |
| Custom subagents | `.claude/agents/*.md`: your own tools, model, prompt |
| Parallel agents | Several independent subagents run together in a single message |
| Workflows | Orchestrate dozens of agents from a script (pipelines like review → verify); only when the user explicitly asks |
| Git worktrees | Each agent in its own worktree, so no conflicts |
| Agent teams / messaging | Send messages to running agents, collect results |
| Option | Detail |
|---|
| /sandbox | Run Bash commands in filesystem + network isolation; fewer permission prompts |
| --dangerously-skip-permissions | = bypassPermissions; only in a disposable container/VM |
| /permissions, deny rules | Block things like .env, secrets, rm -rf |
| Managed settings | Org admin policy that users cannot override |
| Prompt injection guard | Treat instructions in web/file content as data; approval for suspicious commands |
| /security-review | Security review of pending changes |
claude --model opus --permission-mode plan # set model + mode
claude -p "fix the tests" --allowedTools "Bash(npm test:*)" Edit --max-turns 10
claude --add-dir ../shared-lib # access to an extra folder
claude --append-system-prompt "Always write in Hinglish"
claude doctor # installation health check
# Env vars
ANTHROPIC_API_KEY=... # auth with an API key (instead of a subscription)
CLAUDE_CODE_USE_BEDROCK=1 # via Amazon Bedrock
CLAUDE_CODE_USE_VERTEX=1 # via Google Vertex AI
my-plugin/
├── .claude-plugin/plugin.json # name, version, description
├── commands/ # slash commands (.md)
├── agents/ # subagents (.md)
├── skills/<name>/SKILL.md # skills
├── hooks/hooks.json # hooks
└── .mcp.json # MCP servers
# Marketplace = a repo whose .claude-plugin/marketplace.json lists plugins
/plugin marketplace add my-org/claude-plugins
/plugin install my-plugin@my-org
| Surface | Detail |
|---|
| Terminal CLI | Main experience; Windows, macOS, Linux |
| VS Code / JetBrains | Inline diffs, selection context, IDE diagnostics |
| Desktop app | Sessions in a GUI, parallel tasks |
| Web (claude.ai/code) | Work on a repo in a cloud sandbox, can be started from mobile too |
| Remote Control | Control a local session from your phone/browser |
| Slack | Delegate a coding task from a thread |
| Agent SDK | Use Claude Code's harness in your own product |
- /code-review: bug-focused review of the current diff or a PR (with effort levels, apply with --fix).
- /simplify: reuse, simplification, efficiency cleanups.
- /security-review: security issues.
- /init, /fewer-permission-prompts (adds common safe commands to the allowlist), /run (runs the app to verify a change).