---
name: omega-skill-evaluation
description: Use when deciding whether a skill actually improves agent performance. Run paired with-skill/without-skill trials, deterministic tool-call assertions, latency deltas, lift and win-rate gates before promotion.
---

# Empirical skill evaluation

## Procedure

1. Inspect the existing OMEGA capability catalog before adding anything new.
2. Reuse an existing capability when semantics and safety boundaries already match.
3. Add a new capability only when it contributes a distinct executable mechanism.
4. Keep discovery metadata concise; load detailed references only when activated.
5. Verify consequential behavior with tests or observed tool output.
6. Never convert an external project claim into an OMEGA capability claim without local or provider evidence.

See `../../references/ecosystem-research-v14.md` for provenance and `../../references/dedup-policy-v14.md` for canonicalization rules.