SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution

2026-05-18 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

SkillsVote is a lifecycle-governance framework designed for managing Agent Skills, which are defined as an experience schema combining executable scripts with non-executable procedural guidance. The framework addresses challenges in open skill ecosystems, such as redundancy, uneven quality, and environment sensitivity, which can lead to context pollution. SkillsVote profiles a million-scale open-source corpus to assess environment requirements, quality, and verifiability, then synthesizes tasks for verifiable skills. It performs agentic library searches to provide instructional skill context before execution. After execution, SkillsVote decomposes trajectories into skill-linked subtasks, attributes outcomes, and only admits successful, reusable discoveries for evidence-gated updates. Evaluations show offline evolution improves GPT-5.2 on Terminal-Bench 2.0 by up to 7.9 percentage points, and online evolution improves SWE-Bench Pro by up to 2.6 percentage points.

Key takeaway

For research scientists developing long-horizon LLM agents, implementing a governance framework like SkillsVote can significantly enhance agent performance and reusability. You should focus on structuring skill libraries, verifying skill quality, and establishing clear mechanisms for attributing outcomes and updating skills based on successful execution to prevent context pollution and improve agent capabilities without needing model updates.

Key insights

SkillsVote governs agent skill lifecycles, improving LLM agent performance through structured collection, recommendation, and evolution.

Principles

• Agent skills couple executable scripts with non-executable guidance.
• Governed skill libraries improve frozen agents without model updates.
• Control exposure, credit, and preservation for effective skill evolution.

Method

SkillsVote profiles skill corpora, synthesizes verifiable tasks, performs agentic library search, decomposes trajectories, attributes outcomes, and gates updates based on successful, reusable discoveries.

In practice

• Profile open-source skill corpora for quality and verifiability.
• Decompose agent trajectories into skill-linked subtasks.
• Implement evidence-gated updates for reusable skill discoveries.

Topics

• Agent Skills Governance
• LLM Agent Evolution
• SkillsVote Framework
• Experience Schema
• Terminal-Bench 2.0

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks — Agent Skills significantly boost LLM performance, but effectiveness varies, and self-generation is currently ineffective.
• SkillJuror: Measuring How Agent Skill Organization Changes Runtime Behavior — Skill organization profoundly impacts LLM agent runtime behavior and task outcomes, independent of skill content.
• Skills on the Fly: Test-Time Adaptive Skill Synthesis for LLM Agents — Dynamically synthesizing task-specific skills from retrieved trajectories improves LLM agent performance without parameter updates.
• SkillCAT: Contrastive Assessment and Topology-Aware Skill Self-Evolution for LLM Agents — SkillCAT enables LLM agents to self-evolve skills by contrastive assessment and topology-aware execution, boosting performance without training.
• Beyond Prompting: Using Agent Skills in Data Science — Skills package instructions and resources for LLMs, enabling reliable automation of recurring data science workflows.
• A Framework for Evaluating Agentic Skills at Scale — Agent skills significantly enhance LLM behavior and performance, especially for instruction following and making cheaper models competitive.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.