harness-review

To audit overnight runner components in a Claude Code harness project, run harness-review against your curated inventory to classify each item using a four-question rubric, producing per-component verdicts and a candidates-for-review list.

A risk-based rubric for classifying harness components applies four structured questions to each inventory item, evaluating module docstrings to assign a label of load-bearing, experiment-candidate, or likely-deprecated based on its operational risk and utility.

You identify likely-deprecated modules during a harness pruning pass by evaluating their inventory entries and docstring rationale against a four-question rubric, which systematically flags components no longer serving core overnight runner functions.

Yes, you can use harness-review for a Claude Code upgrade cycle without external dependencies. It operates independently to process your inventory components and generate verdicts guiding refinement, pruning, or experimentation decisions.

The best way to classify experiment-candidate components in an overnight runner inventory is applying a structured four-question rubric that reads module docstrings for rationale, systematically evaluating each item's potential for future experimentation versus its current load-bearing status.