SkillGym Released: Automated Pipeline for Training Skill-Use Agents with Verifiable Environments

Summary:SkillGym introduces an automated pipeline for training skill-use agents by constructing verifiable environments, collecting trajectories, and fine-tuning Large Language Models (LLMs). The system crawls skills from the internet, filters those with reproducible workflows, and uses a builder-reviewer pipeline to create difficulty-controlled tasks with reference solutions and executable verifiers. SkillGym has built 6,800 environments and collected 19,000 verified successful trajectories for supervi

---

Key Breakthroughs

SkillGym is an innovative automated tool designed to address the challenges of training agents for skill usage in complex tasks. Its main features include:

• Automated Environment Construction: By crawling skills from the internet and filtering for reproducible workflows, SkillGym can automatically build verifiable training environments.
• Task Difficulty Control: Utilizing a builder-reviewer pipeline, SkillGym generates tasks with controlled difficulty, providing reference solutions and executable verifiers to ensure the reliability of training data.
• Large-Scale Data Collection: SkillGym has built 6,800 environments and collected 19,000 successful trajectories, providing rich data support for supervised fine-tuning.

Technical Highlights

1. Multi-Task Type Support: SkillGym covers four task types, making it adaptable to skill usage needs across different domains.
2. Cross-Model Performance Improvement: The fine-tuned Qwen3.5-9B model outperforms the untrained 397B model on multiple benchmarks, demonstrating SkillGym's universal performance enhancement for LLMs of different sizes and families.
3. Skill Invocation Rate Increase: After training, the rate of agents reading relevant skills increased from 28% to 96%, significantly enhancing their performance in complex tasks.

Industry Impact

SkillGym's release provides a new solution for training agents in complex tasks, with significant implications for the following areas:

• Professional Domain Applications: In fields such as healthcare, finance, and law that require highly specialized knowledge, SkillGym can effectively enhance the professional skill level of agents.
• Multi-Agent Systems: By improving the skill usage capability of individual agents, SkillGym helps to enhance the overall collaboration efficiency of multi-agent systems.
• Edge Computing and Real-Time Applications: The fine-tuned models maintain high performance in resource-constrained environments, providing new possibilities for edge computing and real-time AI applications.

Developer Recommendations

• Stay Updated on SkillGym's Developments: SkillGym is continuously evolving, and developers are advised to stay updated on its latest versions and features to fully leverage its advantages.
• Customize Fine-Tuning for Specific Applications: Customize SkillGym through fine-tuning based on the needs of specific application scenarios to achieve optimal performance.
• Explore Integration with Other Tools: SkillGym can be combined with other AI tools and frameworks to explore its application potential in different fields.

---

Source: Hugging Face Daily Papers (2026-09-29)

— END —

Tags: #SkillGym #Intelligent Agents #Large Language Models #Training Data #Automated Tools