What should a coding-agent task definition include?

Use a task-sized objective that a reviewer can recognize as complete. Avoid broad requests such as improve the system or clean up the code.

Name target paths and constraints before implementation. If the agent discovers a necessary change outside that boundary, require an escalation instead of an unannounced expansion.

The validation command belongs in the task definition, not only in the agent prompt. That makes the definition useful to the next reviewer and the next agent.

A practical checklist for handing work to a coding agent and receiving reviewable implementation evidence back.

Use declared target paths and escalation to prevent hidden scope expansion.

Require executable validation up front and recorded results before coding-agent work can be accepted.