Write Effective Requests

A useful request states the result, relevant context, boundaries, and how the agent should verify completion.

Start with the result

Describe what should be true when the work is complete. Name the affected behavior or component instead of asking the agent to “fix the code” or “improve everything.”

Update the checkout total so an empty cart returns zero. Preserve the existing
function signature and add a test for the empty-cart case.

Add focused context

Attach the file or folder that anchors the request when you know it. Include terminal output or Git changes when they explain a failure or review target.

Do not attach unrelated files merely to provide more context. A focused request is easier to investigate and verify.

State boundaries

Mention interfaces that must remain compatible, files that should not change, commands that require caution, and any expected output format.

Choose an agent whose instructions and permissions match the work. Use a planning agent or Plan Review when the implementation direction needs agreement before editing.

Define verification

Name the test, build, check, or observable behavior that demonstrates completion. If the expected command cannot run in the environment, ask the agent to report that limitation explicitly.

Review the result and evidence before continuing. Refine the same session with specific feedback when the first result does not meet the stated outcome.