Unlock Claude Code’s 15 Hidden Features to Supercharge Your AI Development

On March 30, 2026, Claude Code founder Boris Cherny posted a thread on X titled “I wanted to share a bunch of my favorite hidden and under‑utilized features in Claude Code,” which quickly amassed 1.6 million views. He revealed 15 high‑impact, engineering‑focused features that turn Claude Code from a simple chat‑based coder into a full‑stack, verifiable, isolated, forkable, and schedulable development platform.

1. Mobile App

Claude Code offers native iOS and Android applications. The left‑hand “Code” tab lets you write code, review changes, approve pull requests, and launch new coding tasks directly from your phone, enabling development without a desktop.

2. Cross‑Device Session Transfer

Use claude --teleport or /teleport to pull a cloud‑initiated session (from phone or web) into a local terminal. /remote-control lets a phone or web client take over a running local session. Boris enables “Enable Remote Control for all sessions” in his /config so any session can be hijacked on the fly, allowing tasks started at the office to continue during a commute.

3. Loop / Schedule Automation

The /loop and /schedule commands are the most powerful automation tools. They run tasks at fixed intervals (minimum 5 minutes, up to one week) and keep the session alive for repetitive engineering work. /loop 5m /babysit – automatically handle code reviews and rebase every 5 minutes, shepherding PRs to merge. /loop 30m /slack-feedback – submit PRs and collect Slack feedback every 30 minutes. /loop /post-merge-sweeper – clean up post‑merge review comments. /loop 1h /pr-pruner – close stale PRs hourly.

Note: /loop is session‑scoped and stops when the session ends. Desktop scheduled tasks require the app to stay open; cloud scheduled tasks persist after the computer shuts down.

4. Hooks Lifecycle

Hooks let you attach custom logic at key lifecycle points, making Claude’s behavior deterministic and auditable. Typical hooks include:

SessionStart – load context on each start.

PreToolUse – record all bash commands before tool execution.

PermissionRequest – forward permission dialogs to WhatsApp for remote approval.

Stop – automatically wake Claude when it pauses, ensuring continuous execution.

5. Chrome Extension Verification

The Chrome extension enables Claude to open pages, read DOM state, capture console errors, autofill forms, and run complete user flows directly from CLI or VS Code. This provides self‑validation for frontend code, eliminating the “no feedback” problem.

6. Desktop App Preview Testing

The desktop client can spin up a web server and run previews/tests in an integrated browser, creating a closed “write‑verify‑fix” loop without extra configuration. The Chrome integration is still beta, supporting only Chrome and Edge (not Brave, Arc, or WSL) and requires a direct Anthropic subscription.

7. Session Management

/btw

allows side‑queries that do not pollute the main task context; the query is discarded after answering. /branch forks the current session, enabling experimental work without contaminating the main thread. The --bare flag launches Claude in a minimal mode—skipping hooks, skills, plugins, and automatic project discovery—boosting SDK startup speed up to ten‑fold, though it should not be used when full project context is needed.

8. Parallel Isolation with Git Worktrees

Git Worktrees let multiple Claude processes operate on the same repository in parallel, each with its own isolated worktree. Create a new worktree session with claude -w or enable the “worktree” checkbox in the desktop app. Non‑Git VCS users can implement a WorktreeCreate hook to emulate the same behavior. Proper naming, committing, and merging policies are essential to avoid chaos.

9. Batch Processing

The /batch command distributes large‑scale code changes across dozens or hundreds of worktree agents. Each agent works on its own code copy, making massive migrations, bulk refactors, or rule‑based transformations fast and non‑interfering.

10. Cross‑Repo Authorization

Use --add-dir or /add-dir to grant Claude access to additional project directories, enabling multi‑repo development without launching a separate Claude instance for each repo. Teams can preset directories in settings.json via an additionalDirectories field.

11. Custom Agents

Define reusable agents in .claude/agents/ with system prompts, tool permissions, and model settings. Invoke them via claude --agent=NAME. Example agent types:

Read‑only analysis agent (inspects code without modifying files).

Security audit agent (checks dependencies, permissions, sensitive configs).

Frontend verification agent (prioritises browser‑related tools).

Migration agent (handles large‑scale code replacements).

12. Remote Dispatch

Dispatch lets Claude execute non‑coding tasks (Slack messages, email triage, file management) from a mobile device, separate from Remote Control which only hijacks existing sessions. Dispatch supports both local and cloud scheduled tasks, enabling continuous operation even when the computer is off.

13. Voice Input

The /voice command provides speech‑to‑text coding across platforms: hold the space bar in CLI, click the voice button in the desktop app, or enable system dictation on iOS. This allows most coding to be done by speaking.

14. Prioritized High‑Impact Features

For quick gains, focus on six features: Chrome extension verification, /loop automation, /btw side‑queries, /branch session forking, Git Worktrees for parallel isolation, and custom agents (combined with --add-dir for cross‑repo work).

15. Core Takeaway

By 2026, the decisive factor in AI‑assisted programming is not prompt length but the ability to construct an efficient Claude Code environment—verification, isolation, forking, and scheduling—transforming Claude from a conversational code writer into a production‑grade software engineering system.