Back to feed
arXiv cs.AI·

From Raw Experience to Skill Consumption: A Systematic Study of Model-Generated Agent Skills

Signal
78
Hype
15
In three linesSystematic study of the full lifecycle of model-generated agent skills: extraction, consumption, and transfer. Evaluation framework spanning 5 agentic task domains. Findings: skills beneficial on average but exhibit non-trivial negative transfer; extractor/consumer performance independent of model scale. Introduction of meta-skill to improve quality and reduce negative transfer.
Read source
Your take?
AI AgentsMulti-agentReinforcement learningEvalsPapers

Summary generated by Claude — human-verified