aeon-skill-evals

You can validate AI skill outputs by running them against predefined assertion manifests that check for specific patterns, word counts, and citations to ensure quality and consistency.

Regression detection tracks output changes over time by categorizing results into NEW_FAIL, NEW_PASS, CHRONIC, and STABLE_FAIL statuses to identify when and where quality degraded.

An assertion manifest is a predefined set of criteria used to assess skill outputs, validating that they meet expected rules for word counts, formatting patterns, and required citations.

You can bootstrap a starter assertion manifest by generating it from successful previous runs of the skill, establishing a baseline for future output validation and regression tracking.

Yes, you can generate a starter manifest from successful runs to automatically create initial validation rules, eliminating the need to manually define every assertion from scratch.

Yes, validating outputs against assertion manifests is designed for AI infrastructure management, ensuring consistent quality and detecting regressions across installed skills.