WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution

Noosaga Placements

• Artificial Intelligencesubfield95%
  The paper focuses on AI agent skill evolution, a core topic in artificial intelligence, and is categorized under cs.AI.
  arXiv:2608.27454v1 [cs.AI]We introduce WikiSkill, a framework that co-evolves agent skills with a persistent knowledge base (wiki).
• Neural NLPframework90%
  The method uses LLM-based agents for inference, wiki maintenance, and skill proposal, which are neural NLP systems.
  The Skill Proposer then runs as an autonomous multi-turn ReAct agentYou are a Wiki Maintainer Agent for an LLM skill evolution system.
• Natural Language Processingsubfield90%
  The method relies on LLM-based agents for inference, wiki maintenance, and skill proposal, and is evaluated on NLP-heavy benchmarks.
  The Skill Proposer then runs as an autonomous multi-turn ReAct agentLiveMath ... multiple-choice mathematics competition problemsSealQA ... factual question-answering benchmark
• Attention Mechanisms and Transformersframework70%
  The underlying LLMs are transformer-based, and the study evaluates across different model families and sizes.
  Qwen3.5-4B, Qwen3.5-9B, Qwen3.6-27B, Gemma-4-31B, Gemini 3.5 Flash
• Machine Learningsubfield85%
  The skill evolution process is a form of iterative learning and optimization, with empirical evaluation across models and benchmarks.
  WikiSkill consistently outperforms state-of-the-art skill-evolution methodsskill evolution complements model scaling
• Logical Knowledge Representationframework60%
  The wiki is a structured knowledge base with pattern pages and indexes, representing knowledge in a logical, organized manner.
  The wiki is organized as: - wiki/index.md -- Concise catalog of known patternswiki/patterns/ -- One page per pattern with detailed evidence and analysis
• Knowledge Representation Reasoningsubfield70%
  The wiki serves as a structured knowledge representation that consolidates patterns and insights, supporting reasoning for skill updates.
  maintain a structured knowledge base (wiki) that documents patterns observed during agent executionwiki/index.md -- Concise catalog of known patterns
• Deep Learningsubfield60%
  The agents are based on large language models, and the study examines scaling effects across model sizes.
  larger models generally benefit more from evolved skillsQwen3.5-4B, Qwen3.5-9B, Qwen3.6-27B, Gemma-4-31B, Gemini 3.5 Flash

Abstract

Agent skills package specialized knowledge and workflows into reusable resources that extend AI agent capabilities. Recent work automatically discovers such skills from agent experience, which enables agents to progressively adapt through interaction. However, the insights that guide skill development typically remain scattered across optimization histories, limiting their systematic reuse across iterations. We introduce WikiSkill, a framework that co-evolves agent skills with a persistent knowledge base (wiki). At a high level, WikiSkill separates raw execution experience, accumulated knowledge, and executable skills, while continuously consolidating experience into the wiki, which subsequent skill updates can build on. Across diverse benchmarks and models, WikiSkill consistently outperforms state-of-the-art skill-evolution methods and improves over no-skill baselines in most model-benchmark settings. We find that skill evolution complements model scaling: larger models generally benefit more from evolved skills, while smaller models with skills can outperform substantially larger models without them. We also find that evolved skills transfer effectively across models and model families, and skills evolved by other models can outperform self-evolved skills. Finally, our ablation studies confirm that persistent knowledge accumulation in the wiki is critical for effective skill evolution. These results demonstrate the benefits of systematically accumulating and refining agent experience for developing reusable and transferable skills.

Paper Context