Teleport sessions between devices, automate with /loop and hooks, branch conversations, run parallel agents in worktrees, and cut SDK startup 10x with --bare. Fifteen features most Claude Code users never touch — straight from the person who built the tool.
Boris Cherny built Claude Code. When he lists the features people are sleeping on, it is worth paying attention. In a recent thread he walked through fifteen of them — none exotic, all shipping today, most one command away.
This is a practical digest: what each feature does, when to reach for it, and the one-liner that turns it on.
Go mobile without losing your session
1. The Claude Code mobile app
Claude Code ships a full mobile app. Download it on iOS or Android, tap the Code tab, and you can review diffs, approve PRs, and prompt from the train.
2. Teleport and remote control
Two commands make sessions portable:
claude --teleport(or/teleport) pulls a cloud session down to your local terminal./remote-controllets you drive a local session from your phone or browser.
/config once and stop thinking about it.
Make Claude work while you don't
3. /loop and /schedule
Run Claude on a timer — up to a week of continuous execution. The pattern is: wrap a recurring chore in a skill, then loop it.
/loop 5m /babysit— walks PRs through review and rebase every five minutes./loop 30m /slack-feedback— turns Slack feedback into PRs on a half-hour cadence./loop 1h /pr-pruner— closes stale PRs hourly.
4. Hooks that fire every time
Hooks are scripts triggered by Claude's own events. Unlike a CLAUDE.md instruction, a hook runs every time — no hedging, no forgetting.
SessionStart— preload context from files or an API.PreToolUse— log every bash command before it executes.PermissionRequest— ping your phone on WhatsApp so you can approve from anywhere.Stop— push Claude to keep going if the task isn't actually done.
5. Cowork Dispatch
Dispatch gives you secure remote control of the desktop Claude app. MCP servers, browser, file access — all usable from your phone, with your explicit permission. Useful for triaging Slack and email while you are away from the machine.
Let Claude see what it's building
6. The Chrome extension for frontend work
The rule of thumb: if Claude can see the rendered page, the output gets dramatically better. Ask a human to build a website without a browser and you get the same problem. The Chrome extension is currently more reliable for this than equivalent MCP servers.
7. Desktop app auto-servers
The desktop Claude app will spin up a web server and test it in its built-in browser with zero setup. You can stitch the same loop together in the CLI via the Chrome extension — desktop just gives it to you out of the box.
Flexible conversations
8. Session branching
Want to try a different approach without blowing up the main thread? Two options:
- Run
/branchinside the session. - From the terminal:
claude --resume <session-id> --fork-session.
Return to the original with claude -r <original-session-id>.
9. /btw for side questions
Claude is deep in a task, you need a quick answer. /btw asks without interrupting the agent.
/btw how do you spell "dachshund"?
> Dachshund — German: Dachs (badger) + Hund (dog).The worker keeps going.
Parallelism: the real unlock
10. Git worktrees
Claude Code has first-class worktree support, which means you can run dozens of isolated sessions in the same repo at once.
claude -w— new session in a new worktree.- Or tick the worktree checkbox in the desktop app.
- Non-git VCS? Wire up a
WorktreeCreatehook.
11. /batch for mass changes
/batch clarifies the task, then fans it out across agents in separate worktrees — tens, hundreds, thousands in parallel. The natural fit is large migrations: codemods, framework upgrades, API renames.
Scripting and automation
12. --bare for 10x faster startup
By default, claude -p and the SDKs scan for local CLAUDE.md, settings, and MCP servers on startup. In automation you are already passing --system-prompt, --mcp-config, and --settings — so the scan is pure waste.
claude -p "summarize this codebase" \
--output-format=stream-json \
--verbose \
--bareUp to 10x faster cold start. Boris notes this will become the default in future versions.
13. --add-dir for multi-repo work
Start Claude in one repo, attach another with --add-dir (or /add-dir). This is not just visibility — it grants Claude permission to work in the second repo too. For a permanent setup, add "additionalDirectories" to the project's settings.json.
14. --agent for locked-down roles
Define a specialized agent in .claude/agents/ — its own system prompt, tools, and model — then run:
claude --agent=<agent-name>Good fits: read-only reviewers, security auditors, domain-specific assistants. You get to ship a narrower, safer tool without forking the whole product.
15. /voice
Code by talking. In the CLI, run /voice and hold space to record. The desktop app has a dedicated button. iOS system dictation also works if you want full hands-free drafting.
--bare (for scripts). Those three cover most of the productivity delta.
The bottom line
Most Claude Code users stop at the happy path: type a prompt, read the diff, accept. The features above are what the tool's own author reaches for — automation loops, hooks that fire without fail, parallel agents, portable sessions. None of them require a plugin or a hack. They just require you to try them once.
Keep the full cheatsheet open while you experiment, and skim the hidden features page for the deeper cuts.