COLLEAGUE.SKILL: AI Skill Generation from Expert Knowledge Distillation
A new automated system named COLLEAGUE.SKILL has been developed by researchers to generate AI skills grounded in individual expertise through expert knowledge distillation. This innovation tackles the difficulty of creating LLM agents that encapsulate human judgment, expertise, and interaction styles in a limited manner. By utilizing materials related to a specific individual or role, COLLEAGUE.SKILL generates a skill package that includes two aligned tracks: one for capabilities, mental models, and decision-making processes. This comprehensive workflow transforms diverse data into skills that can be inspected and corrected, addressing the limitations of current memory and persona systems that only capture partial evidence. The research paper can be found on arXiv with the identifier 2605.31264.
Key facts
- COLLEAGUE.SKILL is an automated trace-to-skill distillation system.
- It generates person-grounded AI skills via expert knowledge distillation.
- The system produces a versioned skill package with two tracks: capability and decision-making.
- It addresses the difficulty of building agents with bounded representations of human expertise.
- Existing memory and persona systems capture only fragments of evidence.
- The paper is published on arXiv with ID 2605.31264.
- The system is designed for LLM agents.
- It distills heterogeneous traces into inspectable and correctable skills.
Entities
Institutions
- arXiv