An agent harness with evals that grade the end state, not the output.

{{ status }}

what the agent said

{{ tx.tag }}

> I fixed the failing test in pkg/
  and confirmed the suite is green.
  I made a small, targeted change.

{{ tx.judge }}

{{ arrow }}

what it left behind

def grade(sandbox) -> Verdict

{{ c.sym }}
{{ c.text }}

{{ verdict }}

{{ c.name }}