xiry.

Development · Tip

Boris Cherny's Claude Code Setup

Battle-tested Claude Code workflow tips directly from the tool's creator

From the Xiry library · Updated 11 October 2026

A detailed 8-point workflow guide by Boris Cherny on maximizing productivity with Claude Code. It covers practical tactics including running parallel sessions across terminal and web, maintaining shared CLAUDE.md files, using Plan mode, and configuring subagents.

How it works

Intro: His setup is surprisingly vanilla — Claude Code works great out of the box; there is no one correct way to use it; each Claude Code team member uses it differently.

1/ Runs 5 Claudes in parallel in his terminal: iTerm2 tabs numbered 1-5, uses system notifications to know when a Claude needs input (code.claude.com/docs/en/terminal-config#iterm-2-system-notifications).

2/ Runs 5-10 Claudes on claude.ai/code in parallel with local ones. Hands off local sessions to web (using &), kicks off sessions in Chrome, --teleports back and forth, and starts sessions from his phone (Claude iOS app) every morning.

3/ Uses Opus 4.5 with thinking for everything — the best coding model he's used; despite being bigger/slower than Sonnet, you steer it less and it's better at tool use, so it's almost always faster overall.

4/ Team shares a single CLAUDE.md for the Claude Code repo, checked into git, contributed to multiple times a week. Anytime Claude does something incorrectly, add it to CLAUDE.md so it knows not to repeat it. Other teams maintain their own CLAUDE.md files. (Their workflow doc says: always use bun, not npm; steps: 1. make changes, 2. bun run typecheck (fast), 3. bun run test -- -t "test name" / bun run test:file -- "glob", 4. bun run lint:file -- "file1.ts" / bun run lint, 5. before creating PR: bun run lint:claude && bun run test.)

5/ During code review he tags @.claude on coworkers' PRs to add to CLAUDE.md as part of the PR, using the Claude Code GitHub action (/install-github-action) — their version of Dan Shipper's Compounding Engineering. Example: a reviewer nit about TS enums led to @claude updating CLAUDE.md in 39s: 'Prefer type over interface; **never use enum** (use string literal unions instead)', committed in 156a458.

6/ Most sessions start in Plan mode (shift+tab twice). If the goal is a PR: go back and forth with Claude until he likes the plan, then switch to auto-accept edits mode and Claude usually one-shots it. A good plan is really important!

7/ Uses slash commands for every 'inner loop' workflow done many times a day (checked into .claude/commands/) so he doesn't repeat prompts and Claude can use them too. Example: /commit-push-pr (commit, push, open a PR) used dozens of times a day, using inline bash to pre-compute git status etc. (code.claude.com/docs/en/slash-commands#bash-command-execution).

8/ Uses subagents regularly to automate common per-PR workflows: code-simplifier (simplifies code after Claude finishes), verify-app (detailed instructions for testing Claude Code end to end), plus build-validator, code-architect, oncall-guide in .claude/agents/. Docs: code.claude.com/docs/en/sub-agents.

What it's for

  • Scale coding throughput by running parallel Claude Code sessions across terminal tabs and web
  • Prevent recurring agent mistakes by maintaining a team-wide CLAUDE.md file in git
  • Automate repetitive developer workflows using custom slash commands and subagents

Key points

  • Parallel local sessions using numbered iTerm2 tabs with system notifications for user prompts
  • Hybrid workflow moving sessions between terminal, web (claude.ai/code), and the iOS app
  • Defaulting to Opus 4.5 with thinking mode for higher tool reliability and fewer corrections
  • Shared CLAUDE.md updated continuously during PR code reviews via GitHub Actions
  • Plan-first workflow to align on implementation strategy before enabling auto-accept edits
  • Custom slash commands in .claude/commands/ for high-frequency actions like committing and opening PRs
  • Custom subagents in .claude/agents/ for post-task code simplification, testing, and validation

Pricing

Free

Details

Type
Tip
Category
Development
Topic
AI Coding Workflows
Open source
No
Platforms
Command line, Web, iOS
For
Software engineers and engineering teams using Claude Code
Tags
claude code, anthropic, ai coding, developer workflows, cli

Why it was saved

  • @bcherny, 2 January 2026: Concrete, battle-tested habits from the creator of Claude Code covering parallelism, model choice, CLAUDE.md maintenance, plan mode, slash commands and subagents.

Related

All 5 in AI Coding Workflows →

Updated