The creator of Claude Code has three prompting rules: what, how much effort, how to verify

Boris Cherny, who built Claude Code, posted on October 7 and got 13,000 likes. He's surprised people are surprised by how he prompts. Talk to Claude like a coworker; there's no secret. Back in the Sonnet 3.5 days the prompt mattered a lot. Now what matters is communicating three things.

1. What you want it to do
2. How much effort you want it to spend
3. How it should verify it did the right thing

As a three-line prompt:

Migrate the payment endpoints to the new client.
Small change, so go straight in without a plan, but be thorough running the tests.
Done means: every endpoint uses the new client, the old client is deleted, the test suite passes. Show me the passing output.

Number 2 maps to /effort; number 3 is a check Claude can run itself: tests, a build, a screenshot. The official best-practices doc says the same thing: give Claude a way to verify and a session you watch becomes one you can walk away from.