SkillForge: Co-Evolving Skills and Agents via Dynamic Skill Lifecycles

Key points

• Memory-augmented reinforcement learning strengthens LLM agents' ability to solve complex long-horizon tasks.1
• Skills are one such form of memory, pairing instructions with an applicability condition over task types.1
• Reinforcement learning takes over from this checkpoint, and at each iteration selective retirement, stabilization, and LLM-guided mutation continue to forge the skill library alongside policy optimization.1
• We introduce SkillFurnace, a dataset of 5k+ annotated records bundling retirement-filtered SFT trajectories, evolved skill libraries with fitness annotations, and retirement events with human-annotated failure categories to support research on skill quality and lifecycle management.1

Sources (1)

• [1]SkillForge: Co-Evolving Skills and Agents via Dynamic Skill Lifecycles
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 10:52 AM

  We propose SkillForge, an agentic RL method that compiles and evolves the skill library through a fitness-driven skill lifecycle of trial, active, stable, and retired states, so that the skills and the model co-evolve throughout training.

  Memory-augmented reinforcement learning strengthens LLM agents' ability to solve complex long-horizon tasks.

Extractive summary: sentences quoted from the sources.