SkillFM generates task skills for language model agents without retrieval
SkillFM: Generating Skills for LLM Agents via Latent Flow Matching
Artificial Intelligence
Summary
Some computer programs called large language models (LLMs) can do many tasks, but they often need extra instructions or "skills" to guide them. Usually, these skills come from a fixed list or are learned slowly with trial and error. The authors created SkillFM, a new way to generate these skills automatically and quickly by using a special math method to create skill instructions tailored to each task. SkillFM was tested on tasks like navigating virtual worlds and answering questions, performing better than other methods that rely on looking up stored skills.
What this means in practice
- •For software developers building ai assistants: Create task-specific guidance for AI agents in virtual environments without relying on pre-made skill libraries.
- •For chatbot developers: Generate customized textual steps to improve AI question-answering and online shopping assistants without skill retrieval delays.
Authors
Zuming Zhang, Jie He, Yizhe Zhang, Jeff Z. Pan
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.