Half of Claude’s bad answers come from wrong assumptions it never told you about.
The Prompt
Before you answer, list every assumption you're making
about my situation, my goals, and my constraints.
I'll correct any that are wrong. Then answer.Example
You ask: “Should I use a monorepo?”
Without the assumptions prompt, Claude gives a generic pros/cons list. With it, Claude first says:
I’m assuming: (1) you have 2-5 services, (2) a small team, (3) you’re using TypeScript, (4) you don’t have strong CI/CD infrastructure yet, (5) you want to share types between services.
Now you can say “actually we have 40 services and 12 teams” and get a completely different (and better) answer.
Good for
- Architecture decisions where context matters
- Advice on topics Claude doesn’t have full info about
- Any time Claude’s answer feels generic or off-target
- Debugging wrong recommendations (“ah, you assumed I was using Express, I’m using Fastify”)