COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation

Researchers introduced COLLEAGUE.SKILL, an open-source trace-to-skill system that turns heterogeneous evidence into versioned AI skill packages.

Each package separates capability and bounded behavior tracks, making skills inspectable, editable, and easier to correct or roll back.

The skills can be deployed across agent hosts, supporting portable person- or role-grounded agents.

Public release on Hugging Face, GitHub, and arXiv saw strong adoption, highlighting interest in reusable AI skill generation.