Goal

Use when

• The task has a measurable target, such as a passing test, build, lint check, benchmark, coverage threshold, or expected file output.
• The agent can run a command, inspect an artifact, or otherwise verify whether the target is met.
• The task lacks an explicit goal but needs persistent execution; use a checklist to define and track completion.
• The agent should keep working instead of stopping after a partial attempt.

Don't use when

• The user wants only advice, a plan, or an explanation rather than execution.
• The work cannot proceed without an external decision, approval, credential, or unavailable environment.

Inputs

Capture or infer: