← all entries
S-TIER technique

Boris Cherny's personal CLAUDE.md — plan-first, self-improving rules

keshavsuki · 2026-03-15T15:34:02.000Z · in AI Saves, Vol. 2
Open post ↗

100k ft — What is this?

Boris Cherny's personal CLAUDE.md — plan-first, self-improving rules
techniqueS-tier

50k ft — Why care?

@keshavsuki dissects Boris Cherny's (Anthropic) personal CLAUDE.md: plan-first, subagents for complex work, self-improving lessons.md loop, hard verification standard, elegance demand, autonomous bug fixing, git discipline. High-signal template for CLAUDE.md design.
LLMagents

10k ft — Key takeaways

(no takeaways yet)

5k ft — Tools & techniques

Anthropic, Claude, Claude Code CLAUDE.md authoring, subagent offloading

1k ft — Deep-dive / try it

Comment “boris” for the full file 👇 Boris Cherny is a staff engineer at Anthropic. He shared his personal CLAUDE.md, the rules Claude Code follows every session automatically. Here is what makes it different from what most people write. Plan first. Write the plan before touching any code. If something goes wrong mid-task, stop and re-plan. Never push through. Subagents. Offload complex work, keep the main context clean, throw more compute at hard problems. Self-improvement loop. Every correction becomes a rule in tasks/lessons.md. Next session reads it and applies it. Mistake rate drops over time. Verification standard. Never mark a task complete without proving it works. Run tests, check logs. Ask yourself: would a staff engineer approve this? Demand elegance. For non-trivial chang…
boris-cherny-claude-md
Transcript excerpt
A staff engineer at Anthropic just shared his personal cloud.md file. Here's what's in it and why it actually works. Boris Churny is a staff engineer at Anthropic. He shared his cloud.md file. The rules cloud code follows automatically every single session. Here's what makes it different. Rule 1. Plan mode first. Before touching any code, write the plan. If something goes wrong mid-task, stop and replant it. Never push through. Rule 2. Subagents for hard problems. Offload complex work and keep the main context clean. Throw more compute at it. Rule 3. Self-improvement loop. Processing goes into tasks for slash lessons.md as a rule. Next session reads it and applies it. This drops a mistake rate over time. Rule 4. Prove that it works. Never mark a task as complete without running tests and checking logs. Ask yourself, would a staff engineer approve this? Rule 5. Autonomous bug fixing. When given a bug, just fix it. No hand-holding. Go to the logs, find the root cause, and solve it. Most people treat cloud.md as notes. Boris treats it as a contract. Comment Boris and I'll send you his full file.

Appears in

More from @keshavsuki

Collection: AI Saves, Vol. 2