6arXiv cs.CL (Computation and Language)·
Skill Self-Play: Co-evolving LLM capabilities via structured self-play with dynamic skill routing
Researchers introduce Skill Self-Play (Skill-SP), a reinforcement learning framework that addresses the diversity-vs-verifiability dilemma in LLM self-evolution by using agent skills as a middle ground. The system comprises a proposer, solver, and dynamic skill controller that co-evolve in a continuous loop: the proposer generates tasks conditioned on sampled skills, the solver explores solutions, and the skill controller updates an expanding skill library based on execution feedback. Evaluations on tool-use and reasoning benchmarks show consistent performance gains on capable backbones and recovery for initially misaligned models. Code is released under the Qwen-Applications GitHub organization, suggesting Alibaba/Qwen team involvement.
Related guides (4)
Related events (8)
SkillWeaver: Compositional Skill Routing for LLM Agents via Decompose-Retrieve-Compose
Researchers introduce SkillWeaver, a framework for compositional skill routing in LLM agents that decomposes complex queries into atomic sub-tasks, retrieves matching skills from a large library, and composes an executable DAG plan. The paper formalizes the Compositional Skill Routing problem and introduces CompSkillBench, a benchmark of 300 compositional queries over 2,209 real MCP server skills across 24 categories. A key finding is that task decomposition quality is the primary bottleneck, with standard LLM decomposition reaching only 34.2% category recall; the proposed Iterative Skill-Aware Decomposition (SAD) method improves decomposition accuracy from 51.0% to 67.7% in a single iteration. The framework also reduces context window consumption by over 99% compared to naive skill-stuffing approaches.
SkillComposer: Structured skill composition for LLM agents via constrained autoregressive decoding
A new arXiv preprint introduces SkillComposer, a method that frames skill selection for LLM agents as a structured prediction problem — jointly deciding which skills to activate, how many, and in what order via a constrained autoregressive decoder over skill identifiers. The approach addresses a bottleneck in growing skill libraries where existing retrieval and full-context methods fail to capture the joint nature of skill composition. Evaluated on SkillsBench across two production-grade coding agents (GPT-5.2-Codex and Gemini-3-Pro-Preview), SkillComposer raises pass rates by +23.1 and +18.2 percentage points over no-skill baselines, matching gold-skill retrieval upper bounds at lower prompt-token cost.
SPyCE: Skill-Policy Co-evolution framework for multimodal agents via hierarchical skill libraries
Researchers introduce SPyCE (Skill-Policy Co-evolution), a training framework for multimodal agents that distills reasoning trajectories into a hierarchical skill library co-evolving with the policy during reinforcement learning. The library separates execution skills (local visual operations) from workflow skills (high-level tool orchestration priors), creating a closed loop where better policies yield better skills and vice versa. SPyCE outperforms both RL-based and memory-based baselines across eight benchmarks, suggesting joint skill-policy optimization as a viable paradigm for capable multimodal agents.
MetaEvolve framework trains LLMs to self-evolve via meta-skill reinforcement learning
Researchers introduce MetaEvolve, a framework that cultivates 'meta-skills' such as self-reflection and multi-round refinement in LLMs through a data synthesis pipeline and evolution-aware reinforcement learning. Grounded in coding tasks where program execution provides continuous reward signals, the system trains on evolution trajectories containing programs, fitness scores, and attempt histories. MetaEvolve outperforms the strongest baseline by 10.01% on in-distribution coding benchmarks and 24.12% on out-of-distribution tasks, with a 46.9% relative improvement on open-ended algorithm optimization problems outside the training domain. The work is motivated by AlphaEvolve-style test-time scaling and aims to produce generalizable meta-skills transferable to domains with sparse training signals.
Skill-RM: A unified reward model framework treating evaluation as an agentic skill
Researchers from the Qwen team propose Skill-RM, a framework that reformulates reward modeling as the execution of a reusable 'Reward-Evaluation Skill,' enabling a single model to orchestrate heterogeneous evaluation criteria including rule-based verifiers, ground-truth references, and rubrics. By treating reward computation as a structured agentic task, Skill-RM dynamically selects and aggregates evidence per input rather than relying on static evaluation. Experiments on reward benchmarks and downstream tasks (best-of-N selection, RL) show consistent improvements over traditional judge baselines. The code is publicly released under the Qwen-Applications GitHub organization.
Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution
Role-Agent is a new framework that uses a single LLM simultaneously as both agent and environment, enabling self-bootstrapped co-evolution without external environment feedback. The system has two components: World-In-Agent (WIA), which uses predicted vs. actual state alignment as a process reward, and Agent-In-World (AIW), which reshapes training data by retrieving tasks with similar failure patterns. Experiments across multiple benchmarks show an average performance gain of over 4% over strong baselines. The approach addresses key limitations in LLM agent training: inefficient feedback and static environments.
Double Ratchet: Co-evolving evaluation metrics and skills for self-improving LLM agents
A new arXiv preprint introduces Double Ratchet, a system that co-evolves both evaluation metrics and agent skills in settings where no reliable automatic verifier exists. The metric loop uses evolutionary search over small drawback detectors anchored to a small reference set, while the skill loop uses a lifecycle-managed approach; together they retain 88–110% of the performance lift achievable with ground-truth metrics across code generation (MBPP+), text-to-SQL (Spider 2.0-Snow), and report generation tasks. The paper also addresses safety, showing that anchor discipline and outer audits can catch and repair cases where evolved skills game the rubric. This work directly addresses a core bottleneck in self-improving agent systems: the chicken-and-egg problem of needing a reliable evaluator to improve.
SkillOpt: Systematic Text-Space Optimizer for Self-Evolving Agent Skills
SkillOpt introduces a principled optimization framework for agent skills, treating the skill document as an external trainable state analogous to model weights. A separate optimizer model converts scored rollouts into bounded edits (add/delete/replace) on a skill document, accepting only edits that improve held-out validation scores. Evaluated across six benchmarks, seven target models, and three execution harnesses (direct chat, Codex, Claude Code), SkillOpt achieves best or tied performance on all 52 evaluated cells, lifting GPT-5.5 no-skill accuracy by up to +24.8 points inside the Codex agentic loop. Optimized skill artifacts also transfer across model scales and execution environments without further optimization.