Thinking prompts
Thinking prompts
Short lines you add to any task to change how the agent works: try to prove itself wrong, keep things simple, check its work, hand off cleanly.
- Deeper, not biggerThinking promptUse it before you hand an agent a task you only half know how to describe.
- Prove it wrongThinking promptUse it when an agent's plan sounds right and you can't see why it might be wrong.
- Root cause firstThinking promptUse it the second time the same kind of bug shows up.
- Simpler firstThinking promptUse it when the proposed fix adds a new setting, table or service.
- Let the expert decideThinking promptUse it when an agent keeps asking you technical questions you can't judge.
- Check it independentlyThinking promptUse it before you accept 'tests pass' as proof that something works.
- Spec without the interviewThinking promptThe default way to start a real task. The opposite of /grill-me.
- Reproduce, measure, proveThinking promptUse it in the second session, after a read-only investigation found a real problem.
- Compare three optionsThinking promptUse it for decisions that will be expensive to undo.
- Handoff note before you stopThinking promptUse it before the agent compacts (shrinks its memory of the chat) or a long session ends. /pre-compact does the full version.
- Thorough and boldThinking promptAdd it to any prompt when you want depth and a firm recommendation.