One Brief With A Goal, Boundaries, And Proof
You will finish with a prompt that tells the agent what to change, what not to touch, and how to prove the result works.
Step 1: Choose A Verifiable Outcome
Start with a result a person can see or a test can confirm. “Improve the app” is vague. “Add email validation to the signup form and cover valid and invalid addresses with tests” is verifiable.
Step 2: Give Only Useful Context
Include the stack, relevant directories, current behavior, desired behavior, and existing commands. Avoid a long company history or speculative architecture. The agent needs decision-making context, not a biography.
Step 3: Set Boundaries
- Do not change authentication, billing, deployment, or secrets unless the task requires it.
- Do not add a new dependency without explaining why.
- Stop if the requested behavior conflicts with existing tests or documentation.
- Ask before deleting data, rewriting history, or changing production infrastructure.
Step 4: Define Acceptance Checks
Write the checks before asking for implementation. Include expected behavior, edge cases, test commands, and any accessibility or performance requirement. This converts “done” from an opinion into evidence.
Step 5: Ask For A Reviewable Handoff
Require a summary of changed files, tests run, results, assumptions, and remaining risks. A useful agent response should make human review easier—not merely say the task is complete.
Generate The Full Prompt
Use the free builder to combine these five parts into one copy-ready instruction.
Open the prompt builderBefore You Paste The Prompt
- Remove secrets, customer data, and private credentials.
- Commit or back up the current working state.
- Confirm the agent has only the access it needs.
- Review every change and test result before deployment.
Muse Code status is based on Meta’s Sep 28, 2026 enterprise platform announcement. Meta has not published a Muse Code quickstart as of Sep 29, 2026.