skill-evolution-protocol

Eval-first discipline requires writing a failing evaluation before implementing any skill changes. This testing approach ensures modifications address identified gaps directly while preventing regressions across existing evals.

To apply the minimal-change principle, identify the gap, write a failing eval, implement the smallest possible modification to pass it, and verify the change against all existing evals to ensure no regressions before shipping the update.

You need to split, merge, or deprecate skills when evaluating skill boundaries and lifecycle stages. Governance operations maintain quality and universality by reorganizing overlapping or obsolete skills across projects.

Cross-session learning integrates with skill governance by capturing institutional memory from previous eval cycles. This accumulated knowledge informs future gap identification and ensures continuous improvement loops across multiple sessions.

Yes, you can use this protocol for domain experts, process protocols, and the meta-protocol. It guides users through gap identification, failing-eval creation, and comprehensive verification across the full skill lifecycle.

Enforcing eval-first without dependencies requires strict manual adherence to the improvement loop. Users must independently manage gap identification, failing-eval creation, and cross-skill boundary verification without external tool automation.