aras-skill-builder

To automatically improve underperforming ARAS skills, you can run a skill evaluation that maps paper scores to skills, identifies those with a component score below 7.0, and generates updated SKILL.md content. This process targets skills needing improvement after each research cycle.

Evaluating and updating ARAS skill performance requires performance history data, reviewer feedback, and recent paper scores across skills. This data identifies underperforming skills and generates improved SKILL.md content with an updated performance history.

Versioning during ARAS skill documentation updates involves incrementing the version number and writing the updated SKILL.md to the skill's directory. Older SKILL.md versions are archived to maintain traceability and support A/B evaluation.

The threshold for identifying underperforming ARAS skills is a component score less than 7.0. Skills falling below this score after a research cycle are targeted for SKILL.md content updates and performance history improvements.

Yes, you can archive older SKILL.md versions for traceability. When generating improved content for underperforming ARAS skills, the process archives losing versions to support A/B evaluation and maintain a historical record.

The best way to map paper scores to responsible ARAS skills is using an automated skill evaluation process that analyzes recent scores and reviewer feedback. This identifies underperformers and generates updated documentation for those skills.