Skill Quality

Overview

Improve skills by translating observed failures into small, testable contract changes. A good skill change explains when the behavior applies, what future agents must do, how evals will catch regressions, and what evidence proves the change helped.

This skill governs skill and eval quality decisions. It does not authorize release preparation, version bumps, commits, generated workspace commits, or unrelated package rewrites.

Evidence Intake

Start from the failure signal, not from a preferred rewrite. Collect only the evidence needed for the current decision: