Skip to content

P07: Critic And Capstone

What You Do

Add a critic with iterative refinement and a rubric, score repeated runs, then wire the kept artifacts into harness.py.

Harness Mechanism

Critic, iterative refinement, persisted traces, rubric scoring, and final harness composition.

Open First

Keep

A critic configuration, evaluation rubric, repeated-run evidence, and a runnable harness.py that combines the core P01 to P06 decisions.

Built as a friendly front door for the runnable OpenHands harness lab.