YC CEO Garry Tan open-sourced his Claude Code workflow, hitting 4K+ stars on GitHub in 24 hours.



8 slash commands, each corresponding to a specific role: CEO reviews product direction, Engineering Manager reviews architecture, Staff Engineer identifies production incidents, Release Manager handles one-click deployment, QA auto-captures screenshots to test bugs.

The core idea in one sentence: don't treat AI as an all-purpose assistant—assign it specific roles.

What hits hardest is /plan-ceo-review. You say "add image upload," and it doesn't rush to write code. Instead, it asks back: is this really what you need? Maybe the actual requirement is helping sellers auto-generate product descriptions that actually sell. It pulls "how to do it" back to "what to do."

/review is brutal too—it actively anticipates production disasters: N+1 queries? Race conditions? Failed uploads leaving orphaned files? Think through all the pitfalls before bugs blow up production.

The underlying logic: mixing AI use cases is inferior to specializing each one.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin