Skill-V: Verifiable Self-Evolving Skill Library for Interactive Agents

Key points

• Interactive agents can turn experience into reusable skills, yet existing self-evolving skill libraries primarily improve by accumulating new knowledge.1
• Reliable skill evolution therefore requires not only adding knowledge, but also testing and revising what is already stored.1
• To make stored knowledge testable, we propose representing skills as versioned, falsifiable contracts that link semantic intent to observable behavioral criteria.1
• Applicability-aware filtering reduces incorrect skill invocations, and outcome-grounded revisions correct mis-specified skill boundaries without degrading performance on previously observed evidence.1

Sources (1)

• [1]Skill-V: Verifiable Self-Evolving Skill Library for Interactive Agents
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 11:59 AM

  We introduce Skill-V, a verifiable self-evolving skill library.

  Interactive agents can turn experience into reusable skills, yet existing self-evolving skill libraries primarily improve by accumulating new knowledge.

Extractive summary: sentences quoted from the sources.