Harness Config Optimizer — Proposer Skill

Optimize Claude harness configurations by inspecting failure patterns in execution traces to identify dominant issues and apply the smallest effective changes to prompts, hooks, tools, or permissions. The updated config is written to a staging path for evaluation.

Diagnose recurring Claude hook failures by analyzing execution traces and candidate results to detect failure patterns, which identifies the dominant issue and generates a generalizable rule for the smallest effective harness fix.

Yes, you can use execution traces to apply minimal changes to Claude prompts and permissions by identifying the dominant failure issue and modifying the specific harness logic required to resolve it.

Run the outer-loop workflow to automatically generate a new harness configuration in staging by analyzing traces, then evaluate its impact on the benchmark to iteratively improve build_options.

Automated harness optimization should be avoided when you lack sufficient execution traces or candidate results, as the process relies on inspecting failure patterns to identify dominant issues and apply the smallest effective changes.

No, the Harness Config Optimizer does not require any dependencies to run, as it directly inspects execution traces and failure patterns to apply minimal changes to Claude harness configurations.