SkillFM: Generating Skills for LLM Agents via Latent Flow Matching

2026-09-30

AI Summary

This study addresses the reliance of LLM agents on manually curated skill libraries and the delayed feedback inherent in reinforcement learning by proposing SkillFM, a novel generative framework. The method integrates a continuous latent-space encoder-decoder with an improved MeanFlow model to directly synthesize task-conditioned textual skills via latent flow matching. This enables single-step sampling generation during inference, guiding frozen downstream agents without retrieval. Evaluated on the ALFWorld and Search-QA benchmarks, the proposed approach achieves state-of-the-art performance among vector-based skill methods, validating the effectiveness of the latent generative paradigm for agent skill acquisition.

Abstract

Textual skills provide reusable guidance for large language model agents, but existing approaches often rely on manually curated skill banks or reinforcement learning with indirect and delayed feedback. We introduce SkillFM (Skill Flow Matching), a generative framework that synthesizes task-conditioned textual skills directly without test-time skill retrieval. Our framework combines a codec for encoding and reconstructing textual skills in a continuous latent space with a conditional flow model trained using improved MeanFlow. At inference time, the learned velocity field enables single-step latent sampling, and an LLM-based decoder converts the sampled representation into textual guidance for a frozen downstream agent. We evaluate the framework on embodied tasks, question answering, and web shopping. On ALFWorld and Search-QA, our method achieves the best overall performance among the compared vector-based skill approaches. Our analyses further demonstrate that latent skill generation is an effective alternative to retrieval-based skill augmentation. Our code and training skill libraries are available at https://github.com/lulushang999/SkillFM.

Problem

• LLM agents
• skill generation
• textual skills
• latent flow matching

Innovation

• Latent Flow Matching
• Skill Generation
• MeanFlow
• Continuous Latent Space
• LLM Agents