Skill Evals Optimize

Triage failed eval cases using the steering guide, apply limited fixes, and retest with a strict iteration cap.

Inputs

• Results root: evals/skill-loading/.tmp/opencode-eval-results
• Max optimization iterations: 2
• Steering guide: evals/skill-loading/docs/skill-optimization-steering.md

Workflow

1. Locate latest results and list failed cases

   bash scripts/list-fails.sh