os-skill-improvement

Agent skill routing accuracy is improved by running baseline evaluations, applying focused patches, and refactoring skills based on evaluation results using the RED-GREEN-REFACTOR cycle.

The RED-GREEN-REFACTOR cycle for agent skill development is a continuous improvement process: RED runs baselines to observe failures, GREEN applies patches, and REFACTOR closes loopholes until evaluation scores meet thresholds.

To fix low agent skill scores, use os-eval-runner as an objective evaluation gate to verify focused patches during the GREEN phase and continue refactoring until the target score threshold is met.

Yes, os-eval-runner is a required dependency for continuous agent skill improvement, functioning as the objective evaluation gate to verify patches and measure refactoring progress against target thresholds.

Trigger a skill improvement cycle when agent routing accuracy is low, trigger descriptions need sharpening, or baseline evaluation scores from os-eval-runner fall below your target threshold.

The best way to sharpen trigger descriptions is applying focused patches during the GREEN phase and refactoring to close loopholes, verifying changes against baseline evaluation data until scores pass.