15 Hidden Claude Code Features, Straight From the Person Who Built It

Boris Cherny created Claude Code. On March 29, 2026, he posted a thread on X walking through 15 features he uses daily that most people don’t know about. The thread racked up over 100K views in under an hour.

Cherny works at Anthropic and has been building Claude Code since its early days. When the person who made the tool says “here’s what you’re missing,” it’s worth paying attention.

Here’s every tip from the thread, with added context.

Source: Boris Cherny on X, March 29 2026

1. Claude Code Has a Mobile App

Most people think of Claude Code as a terminal tool. It also runs on your phone. Open the Claude app on iOS or Android, tap the Code tab on the left side, and you have a full Claude Code session in your pocket.

Cherny says he writes a lot of his code from the iOS app. For quick changes or reviewing what an agent did while you’re away from your desk, it removes the need to open a laptop.

2. Move Sessions Between Devices With Teleport and Remote Control

You can pick up a Claude Code session on a different device without losing context. Two ways to do it:

claude --teleport (or /teleport): Moves a cloud session to your local terminal.
/remote-control: Lets you control a locally running session from your phone or the web.

Cherny keeps “Enable Remote Control for all sessions” turned on in his /config. That way any session he starts is accessible from anywhere.

Docs: Remote Control

3. Schedule Recurring Tasks With /loop and /schedule

These two commands let Claude run automatically at set intervals, for up to a week at a time. Cherny calls them two of the most powerful features in Claude Code.

Here’s what he runs:

/loop 5m /babysit to auto-address code review, auto-rebase, and shepherd PRs to production
/loop 30m /slack-feedback to automatically put up PRs for Slack feedback every 30 minutes
/loop /post-merge-sweeper to put up PRs addressing code review comments he missed
/loop 1h /pr-pruner to close out stale and no longer necessary PRs

The pattern: turn a workflow into a skill (a custom slash command), then put it on a loop. The skill does the thinking; the loop does the scheduling.

Docs: Run prompts on a schedule

4. Use Hooks for Deterministic Lifecycle Logic

Hooks let you run your own code at specific points in the Claude Code agent lifecycle. They’re deterministic, meaning they always run when their trigger fires.

Cherny lists four use cases:

SessionStart: Dynamically load context each time you start Claude
PreToolUse: Log every bash command the model runs
PermissionRequest: Route permission prompts to WhatsApp for remote approval
Stop: Poke Claude to keep going whenever it stops

That WhatsApp routing tip is worth highlighting. If you’re running agents overnight or unattended, you can have sensitive permission requests forwarded to your phone so you approve or deny them without sitting at your computer.

Docs: Hooks

5. Cowork Dispatch for Non-Coding Tasks

Dispatch is a remote control for the Claude Desktop app. It can use your MCPs, browser, and computer (with your permission) to do things on your machine when you’re not there.

Cherny uses it every day to catch up on Slack, manage emails, handle files, and run tasks on his laptop when he’s away. His framing: when he’s not coding, he’s dispatching.

Docs: Cowork Dispatch

6. The Chrome Extension for Frontend Work

Cherny’s tip here is simple but easy to miss. The most important thing you can do for Claude Code is give it a way to verify its output.

His analogy: if you ask an engineer to build a website but they can’t use a browser, the result probably won’t look great. Give them a browser and they’ll iterate until it does. Same applies to Claude.

The Chrome extension connects Claude Code to your browser so it can see what the page actually looks like, check console logs, and test its own changes. Cherny uses it every time he works on web code and says it tends to work more reliably than similar MCP setups.

Docs: Use Claude Code with Chrome

7. Desktop App for Auto-Starting Web Servers

The Claude Desktop app can automatically start your web server and test it in a built-in browser. You don’t need to set anything up. It just works.

If you prefer the CLI or VS Code, you can get similar functionality through the Chrome extension. But the Desktop app bundles it all together with zero configuration.

Docs: Claude Code Desktop

8. Fork Sessions With /branch

Sometimes you want to take a conversation in a different direction without losing the original. Two ways to fork:

Run /branch from inside your session
From the CLI: claude --resume <session-id> --fork-session

This is useful when you want to explore an alternative approach without blowing up your current working session.

9. Side Queries With /btw

While Claude is working on something, you can ask it a quick unrelated question using /btw. It answers your question without interrupting the main task.

Cherny says he uses this constantly. Need to check how to spell something, look up a function signature, or ask a quick question? /btw handles it without derailing what the agent is doing.

10. Git Worktrees for Parallel Work

Claude Code has deep support for git worktrees. Worktrees let you have multiple checkouts of the same repository, which means you can run many Claude sessions in parallel on different branches without them stepping on each other.

Cherny says he has dozens of Claudes running at all times. To start a new session in a worktree, use claude -w or check the “worktree” checkbox in the Desktop app. For non-git version control, you can use the WorktreeCreate hook to add your own logic.

11. Fan Out Massive Changesets With /batch

/batch interviews you about what you want to do, then spins up as many worktree agents as needed to get it done. Dozens, hundreds, even thousands of parallel agents working on the same codebase.

The main use case: large code migrations and other parallelizable work. If you’ve ever needed to update an import across 500 files or migrate an API across a monorepo, this is the tool for that.

12. Speed Up SDK Startup With —bare

By default, running claude -p (or using the TypeScript/Python SDKs) searches for local CLAUDE.md files, settings, and MCPs. For non-interactive usage like scripts and pipelines, that’s wasted time.

The --bare flag skips all of that and can speed up startup by up to 10x. You then explicitly specify what to load using --system-prompt, --mcp-config, --settings, etc.

Cherny notes this was a design oversight from when they first built the SDK. A future version will make --bare the default for non-interactive mode.

13. Access Multiple Repos With —add-dir

When working across multiple repositories, use --add-dir (or /add-dir during a session) to give Claude visibility and permissions in additional folders.

This does two things: tells Claude about the other repo and grants it permission to read and write files there. For teams, you can add additionalDirectories to your settings.json so the extra folders load automatically.

Docs: CLI reference

14. Custom Agents With —agent

Custom agents let you give Claude Code a specific system prompt and a restricted set of tools. You define an agent file in .claude/agents, then run claude --agent=<name>.

Cherny says this is a powerful primitive that often gets overlooked. You could make a read-only agent that can only use the Read tool, a code review agent with specific instructions, or a deployment agent with access to only the tools it needs.

Docs: Sub-agents

15. Voice Input With /voice

Run /voice in the CLI and hold the space bar to talk instead of type. The Desktop app has a dedicated voice button. On iOS, you can enable dictation in your device settings.

Cherny says he does most of his coding by speaking to Claude rather than typing. For people who think faster than they type (most people), this can significantly change the speed of interaction.

The Takeaway

A few patterns stand out across these 15 tips. First, Claude Code is clearly designed to run across devices and contexts, not just in a terminal on your laptop. Second, the automation layer (hooks, loops, schedules, batch) is where the power-user features live. Third, giving Claude ways to verify its own output (Chrome extension, built-in browser) makes a real difference in quality.

Cherny ended the thread asking what other underrated features people use. Worth following him at @bcherny if you want to catch the next batch.