7 rules for prompts that get better results

The agent is smart, but it can’t read your mind. Every missing detail gets guessed, and every wrong guess means another round and more credit. These seven rules will make most of your requests land the first time.

1. Say "what" and "why"

Instead of "fix the button", write: "Make the Order now button on the homepage bigger, because mobile visitors miss it." The reason helps the agent make good calls on details you didn’t mention.

2. Name the place

A file, page or component name saves the agent search time: "in Hero.tsx" or "on the settings page".

3. Share the references you have

The full error message, a screenshot, a docs link, or "the colors are in theme.css". Whatever you already know, hand it over instead of making the agent look for it.

4. State the constraints

• "No new libraries."
• "Don’t change the API shape."
• "It must work in older browsers."

5. Ask for a plan on big tasks

For large changes start with: "Propose a plan to build the coupon system without writing code." Review the plan, adjust it, then ask for the implementation. For very long tasks use /goal so the agent keeps going across several rounds.

6. Ask it to verify

End your request with "run the tests" or "run the build and make sure it passes". That way the agent catches and fixes its own mistakes before it reports back.

7. One task per chat

Short, focused chats are more accurate and cheaper. When a task is done, start a new chat for the next one. If a chat gets long, use /compact to summarize it and continue with less context.

Before and after