Co-Evolving LLM Decision and Skill Bank Agents for Long-Horizon Tasks
For LLM-based agents in long-horizon interactive environments, COSPLAY addresses the bottleneck of skill discovery and reuse, showing substantial gains over existing methods.
COSPRAY, a co-evolution framework for LLM agents, improves long-horizon decision making by having a decision agent retrieve skills from a learnable skill bank while a skill pipeline discovers reusable skills from rollouts. It achieves over 25.1% average reward improvement against frontier LLM baselines on single-player games.
Long horizon interactive environments are a testbed for evaluating agents skill usage abilities. These environments demand multi step reasoning, the chaining of multiple skills over many timesteps, and robust decision making under delayed rewards and partial observability. Games are a good testbed for evaluating agent skill usage in environments. Large Language Models (LLMs) offer a promising alternative as game playing agents, but they often struggle with consistent long horizon decision making because they lack a mechanism to discover, retain, and reuse structured skills across episodes. We present COSPLAY, a co evolution framework in which an LLM decision agent retrieves skills from a learnable skill bank to guide action taking, while an agent managed skill pipeline discovers reusable skills from the agents unlabeled rollouts to form a skill bank. Our framework improves both the decision agent to learn better skill retrieval and action generation, while the skill bank agent continually extracts, refines, and updates skills together with their contracts. Experiments across six game environments show that COSPLAY with an 8B base model achieves over 25.1 percent average reward improvement against four frontier LLM baselines on single player game benchmarks while remaining competitive on multi player social reasoning games.