10 Claude Code Tips from Boris Cherny

Tips from Boris Cherny, creator of Claude Code, filtered for product teams.

1. Run Multiple Sessions in Parallel

Run 5 sessions simultaneously in terminal tabs (numbered 1–5) with system notifications enabled. Each session has its own context and memory. Tasks must not touch the same source files.

Enable notifications so Claude can alert you when waiting for input.

2. Use /voice for Voice Input

/voice enables voice input mode. Useful for complex tasks — when speaking, you naturally focus less on crafting the perfect prompt and more on describing intent.

3. Use /teleport to Jump Between Devices

Move sessions between mobile, web, and desktop:

claude --teleport    # continue a cloud session on your machine
/teleport            # same, from inside a session
/remote-control      # control a local session from phone/web

Configure remote access in /config → "Enable Remote Control for all sessions".

CLAUDE.md acts as Claude's memory for the project. Boris's team adds it to Git and the whole team contributes to it multiple times a week — keeps results consistent across team members.

See The Complete Guide to CLAUDE.md for writing guidance.

5. Use --add-dir for Multi-Repo Access

Start Claude in one repo and add folders from another:

claude --add-dir /path/to/other/repo

Grants permissions to work in that directory. Useful for shared brand language or shared libraries across projects.

6. Use /loop and /schedule for Recurring Tasks

/loop — repeat a task at intervals inside the current session:

/loop 60 Check new user feedback in /data/reviews.txt and summarize key UX issues

/schedule — trigger-based remote agent (not a simple timer):

/schedule create
name: "ux-daily-insights"
trigger: "0 9 * * *"
task: "Analyze /data/analytics.json and /research/feedback.md. Save report to /reports/daily.md"

Difference: /loop runs inside your session; /schedule runs outside it on a cron trigger.

7. Use /btw for Side Queries

Ask Claude a quick question while it's working on the primary task, without interrupting the main flow.

8. Use /batch to Fan Out Across Agents

Parallelizes work across many agents:

/batch
Audit all screens in /ui-screens directory for:
- usability issues
- accessibility problems
- consistency with design system

Claude splits the problem → spawns agents per bucket → executes in parallel → merges results. Claude will ask for clarifying details (file format, design system location, severity scale, output format) — better answers produce better decomposition.

9. Use Hooks in the Agent Lifecycle

Hook events:

10. Fork Sessions

Fork = create a new session from the current state, keeping context, files, decisions, and history — then go in a different direction.

/branch                                      # from inside a session
claude --resume <session-id> --fork-session  # from CLI
Powered by Forestry.md