Claude Code Hidden Features
Most Claude Code users type a prompt, get a response, and move on. But the CLI is packed with built-in features that change how you work with it. Here are the ones I keep coming back to.
/statusline — Know What's Happening at a Glance
Have you ever spent time trying to figure out your usage statistics? This skill lets you keep these figures in the status line, showing you what Claude is doing in real time: current working directory, git branch, token usage, and session info. It's off by default, which means most people never see it.
Run /statusline to configure it. You can choose between a minimal display or a detailed one that shows model, token count, and active tool calls. Once you turn it on, you'll wonder how you worked without it — especially during long tasks where you want to know if Claude is still thinking or stuck waiting on a tool.
/remote-control — Your Terminal Session, From Any Device
This one changes how you think about where you work. Remote control lets you continue a Claude Code session running on your local machine from any browser or the Claude mobile app. Your filesystem, MCP servers, tools, and project config all stay local — you're just steering from somewhere else.
The use case: you kick off a long refactor at your desk, then move to the couch and keep directing it from your phone. Or you're pair-programming and want a colleague to see your session in their browser without screen sharing.
Three ways to start it:
claude remote-control— dedicated server mode that displays a session URL and QR code. Supports multiple concurrent sessions.claude --remote-control "My Project"— interactive session in your terminal that's also available remotely./remote-controlfrom inside an existing session — enables it on the fly with your full conversation history intact.
Connect from another device by opening the session URL, scanning the QR code with the Claude app, or finding the session by name in claude.ai/code. Everything stays encrypted over TLS through the Anthropic API — your machine never opens inbound ports.
The key distinction from Claude Code on the web: remote control extends your local environment to another screen. Claude Code on the web runs in the cloud with no access to your files. Remote control gives you both.
/compact — Reclaim Your Context Window
Long conversations eat context. /compact compresses your conversation history so you can keep working without starting a new session. The useful trick: you can pass it a prompt to guide what gets preserved.
/compact focus on the auth refactor, drop the earlier debugging
This is especially valuable during multi-hour sessions where early exploration becomes irrelevant to the current task.
/vim — Vim Keybindings in the Input
If you live in vim, this is a one-command quality-of-life upgrade. /vim enables vim-style keybindings in the Claude Code input field. Normal mode, insert mode, visual mode — it all works. No more reaching for arrow keys.
/terminal-setup — Shell Integration Most People Skip
During install, Claude Code offers to set up shell integration and most people click past it. /terminal-setup lets you revisit that. It configures your prompt to show Claude Code status, sets up the terminal title bar, and enables better context passing between your shell and Claude.
/init — Let Claude Write Your CLAUDE.md
Every project benefits from a CLAUDE.md file that tells Claude about your codebase conventions, structure, and constraints. Writing one from scratch is tedious. /init scans your project and generates a solid first draft automatically. It picks up on your tech stack, file structure, and patterns.
Start here if you haven't set up a CLAUDE.md yet. Edit the output, don't just accept it blindly — but it's a much better starting point than a blank file.
/review — Code Review Without a PR
/review runs a code review on your current uncommitted changes. It's useful as a pre-commit sanity check: catch obvious issues before they hit CI or a human reviewer.
I use this as a habit before running /commit. It catches things like forgotten debug statements, inconsistent error handling, or tests that don't assert anything meaningful.
The ! Prefix — Run Shell Commands in Context
Not a slash command, but too useful to skip. Prefix any shell command with ! and its output lands directly in your conversation:
! git log --oneline -10
Now Claude can see your recent commits and reason about them. Works with any command — ! curl, ! cat, ! docker ps. It bridges the gap between "Claude can read files" and "Claude understands my current environment."
Shift+Tab — Toggle Auto-Accept
By default, Claude asks permission before running tools. Shift+Tab toggles auto-accept mode so it can work autonomously. Turn it on when you trust the task (running tests, reading files), turn it off when Claude is about to do something irreversible.
--dangerously-skip-permissions — Full Autonomy Mode
If Shift+Tab is giving Claude a longer leash, --dangerously-skip-permissions is taking the leash off entirely. Launch a session with:
claude --dangerously-skip-permissions
Claude will execute every tool — file writes, shell commands, everything — without asking. No confirmation prompts, no permission gates. It's designed for trusted, automated workflows: CI pipelines, batch scripts, or situations where you've already reviewed the plan and just want Claude to execute.
The name is intentional. This flag skips all safety prompts, so use it when you understand exactly what Claude will do and you're prepared for it to act without guardrails. Don't use it for exploratory work where you'd want to review each step. Using well established open sourced skills and plugins should generally be ok, but use at your own risk.
Wrapping Up
None of these are hidden behind flags or experimental settings — they ship with every Claude Code install. The pattern I've noticed: the features that save the most time are the ones that reduce friction between you and Claude, whether that's context management (/compact), environment awareness (/terminal-setup, !), or workflow integration (remote control, /statusline).
Try one or two this week. You'll be surprised how much smoother the experience gets.