Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills

2026-07-24

AI Summary

This work addresses the fundamental tension between task diversity and verification reliability in the self-evolution of large language models by proposing the Skill Self-Play framework. The framework introduces skills as a structured intermediate layer, enabling skill-guided task generation and dynamic routing through a reinforcement learning–driven self-play loop among a proposer, a solver, and a dynamic skill controller. It supports scalable construction of a skill library and continuously refines skills based on execution feedback. Experimental results demonstrate that the framework significantly enhances the performance of strong base models on tool-use and reasoning benchmarks and effectively mitigates capability deficiencies arising from initial misalignment.

Abstract

LLM training is shifting from manual design and annotation to interaction-driven self-evolution. However, existing self-evolutionary methods face a fundamental dilemma between task diversity and verification reliability: environment-bound methods obtain precise feedback but confine learning to narrow domains, while open-ended self-generation broadens the task space but lacks reliable verification, allowing misleading rewards to pollute the training loop. We identify agent skills as a powerful middle ground to reconcile this tension: each skill ensures deep, verifiable execution in a specific scenario, while dynamic routing across skills maintains open-ended task variety. Leveraging this insight, we introduce Skill Self-Play (Skill-SP), a co-evolutionary framework comprising a proposer, a solver, and a dynamic skill controller. Orchestrated via a reinforcement learning loop, these components co-evolve in a continuous self-play loop: the proposer generates challenging tasks conditioned on dynamically sampled skills; the solver explores candidate solutions to push its capability boundaries; and the skill controller collects execution feedback to update and expand the skill library. This interactive co-evolution effectively bridges the gap between structured verification and open-ended exploration. Empirical evaluations on tool-use and reasoning benchmarks demonstrate that Skill-SP, serving as a robust evolution engine, consistently pushes the performance ceiling of competent backbones while catalyzing striking turnarounds for initially misaligned models. Our code is available at https://github.com/Qwen-Applications/skill-self-play.

Problem

• LLM self-evolution
• task diversity
• verification reliability
• open-ended generation
• environment-bound learning

Innovation

• Skill Self-Play
• co-evolution
• self-play
• dynamic skill routing
• reinforcement learning