Claude Code Hidden Features – 15 Power Tips from Boris Cherny (Anthropic)

Claude Code is growing up

Boris Cherny, engineer at Anthropic, recently shared 15 features that most Claude Code users don't know about. No marketing fluff – real power-user capabilities that fundamentally change how you work with the tool.

We use Claude Code daily as our primary engineering tool. Most of these features aren't new to us – but the way Anthropic brings them together shows a clear direction: Claude Code is evolving from a CLI chat into a Developer Operating System.

Here are the features that actually make a difference – with our take.

Voice-First Coding with /voice

Hold spacebar, talk, release. Boris says he now codes mostly by voice. Sounds absurd – but works surprisingly well for:

• High-level instructions: "Refactor the auth logic, extract the token refresh into its own hook"
• Code reviews: "Explain what this function does and where the edge cases are"
• Quick fixes: "Add error handling for the API call on line 42"

Voice isn't optimal for every context. But for the 80% of interactions that are natural language, it's simply faster than typing.

Mobile App – Claude Code on your phone

Full Claude Code on iOS and Android. Review PRs, fix bugs, push commits – from your smartphone. Sounds like a gimmick, but in practice it's gold:

• Sunday evening, Slack notification: production bug. Quickly commit the fix from your phone without opening the laptop.
• On the train between meetings: go through code reviews and give feedback.
• Pair programming with the team while you're on the move.

Teleport Sessions

claude --teleport or simply /teleport in the chat. Start a session on desktop, continue on mobile – or vice versa. The complete context travels with you.

This isn't a feature, it's an architecture decision. Anthropic treats sessions as portable units, not device-bound processes. It's the same thinking behind their Brain ↔ Hands Decoupling.

/loop and /schedule – Automated Workflows

This is where it gets exciting for teams:

• /loop 5m /babysit – Runs babysit mode every 5 minutes. Perfect for CI monitoring or test watching.
• /schedule 9am daily /standup – Recurring workflows. Every morning at 9, Claude generates your standup update based on git history and open issues.

This isn't a toy. This is Agentic Automation – Claude Code as a continuously running agent, not an on-demand tool.

/batch – 50 Agents in parallel

The game-changer for large codebases. /batch spins up to 50 parallel agents working on different files simultaneously. Use cases:

• Codebase-wide refactoring: Migrate all API calls to a new schema
• Test generation: Write unit tests for every file in a directory
• Dependency updates: Fix breaking changes in 50 files at once

This is the moment when "Agent" in "Agentic Engineering" stops being a buzzword. 50 parallel agents, coordinated through a single command – that's Agentic Engineering in practice.

Git Worktrees – Parallel Branches

claude -w or simply check the worktree box. Claude works in a separate Git worktree – your main branch stays untouched.

Why this matters: No more context switches. You can have Claude work on an experimental feature while you continue working in the main branch. Perfect isolation, zero overhead.

--bare – 10x Faster Startup

claude --bare skips the automatic config search and loads minimally. When you know what you want, you save the startup time.

Sounds like micro-optimization – but it's relevant when you start Claude Code dozens of times a day. Seconds add up.

Hooks – Lifecycle Events

This is where the OS ambition shows most clearly. Hooks let you inject logic at specific lifecycle moments:

• SessionStart: Automatically load context (project README, current sprint goals, team conventions)
• PreToolUse: Log every command before it executes
• PermissionRequest: Auto-approve safe tools, always confirm critical ones

That's plugin architecture. That's an extension system. That's an Operating System orchestrating developer workflows – not just answering questions.

/branch – Fork Sessions

/branch or claude --resume ID --fork-session creates a copy of your current session. Perfect for:

• "What if" exploration without risking the main session
• Parallel solution approaches: Try two different architectures, keep the better one
• Rollback safety: Fork the session before a risky refactoring

/btw – Side Queries

Side queries while your main agent keeps working. "Btw, what's the Jest matcher syntax for async again?" – without interrupting the running process.

Sounds small. But it's the difference between a tool you operate and a tool that works with you.

The Big Picture: Claude Code as Agent OS

These features individually are nice. Taken together, they tell a story:

Anthropic isn't building a better chat interface. They're building an Operating System for developer work.

• Voice + Mobile + Teleport = Device-independent work environment
• Loop + Schedule + Batch = Autonomous agent orchestration
• Hooks + Branch + Worktrees = Extensible developer platform

This fits exactly into the trajectory we described in our Claude Code GTM Layer article: Claude Code isn't Anthropic's developer tool – it's their go-to-market strategy for the enterprise market.

And it fits into the Agentic Coding Landscape: While Cursor and Copilot optimize the editor, Anthropic is building the infrastructure layer beneath it.

What this means for you

If you're still using Claude Code like a chat tool – question in, answer out – you're leaving 90% of the potential on the table.

Our recommendations:

1. Set up Hooks – Feed SessionStart with your project context
2. Enable Worktrees – Isolated branches for every major task
3. Use /batch for refactoring – Never manually touch 50 files again
4. Try Voice – Seriously. It's faster than you think.

Want to use Claude Code as a strategic engineering tool, not just a coding assistant? Talk to us – we help teams go from "AI-assisted" to "AI-first."