OpenSkill: Open-World Self-Evolution for LLM Agents

2026-06-04

AI Summary

This work addresses the challenge of enabling large language model (LLM) agents to autonomously evolve in open-world settings where supervision signals are scarce. To this end, the authors propose OpenSkill, a framework that extracts open-domain knowledge from documents, code repositories, and the web to synthesize transferable skills. These skills are refined through unsupervised training within self-constructed virtual task environments and evaluated using only task-specific prompts—without any ground-truth supervision. OpenSkill is the first approach to achieve agent self-evolution without requiring labeled data for target tasks. Its self-built validator aligns closely with real-world outcomes, and the learned skills generalize across different models. Experiments demonstrate that OpenSkill achieves state-of-the-art automatic pass rates across three benchmarks and two types of agents, validating its effectiveness and generalization under strictly unsupervised conditions.

Abstract

Self-evolving agents requires adaptation after deployment, but existing approaches assume a usable learning loop, such as curated skills, successful trajectories, or verifier signals. Real open-world deployments may provide none of these, offering only a task prompt. In this work, we study open-world self-evolution, where an agent must build both its skills and its own verification signals from scratch, using open-world resources but no target-task supervision. We propose OpenSkill, a framework that bootstraps this loop: it acquires grounded knowledge and verification anchors from documentation, repositories, and the web, synthesizes them into transferable skills, and refines those skills against self-built virtual tasks grounded in the anchors rather than in target answers. The open world thus supplies both the knowledge to be learned and a supervision-independent practice environment, with target-task supervision reserved for final evaluation. Across three benchmarks and two target agents, OpenSkill attains the best automated pass rate while satisfying the no-supervision constraint. Analysis shows its skills transfer across models without model-specific adaptation, and its self-built verifier aligns with ground-truth outcomes despite never accessing them.

Problem

• open-world
• self-evolution
• LLM agents
• zero-supervision
• skill acquisition

Innovation

• open-world self-evolution
• self-supervised skill acquisition
• verification-free learning
• transferable LLM skills
• virtual task synthesis