SkillClaw: Collective Skill Evolution with Agentic Evolver (April 2026)

Free

Cross-user trajectories continuously aggregated and refined by autonomous evolver into shared skill repository — collective skill evolution in multi-user agent ecosystems; 142 HF likes

FreeFree tier

Type

Open Source

Visit Website GitHub (40 stars)

About SkillClaw: Collective Skill Evolution with Agentic Evolver (April 2026)

SkillClaw is a framework for collective skill evolution in multi-user LLM agent ecosystems. It continuously aggregates cross-user trajectories and uses an autonomous evolver to identify recurring behavioral patterns, translating them into updates to the skill set by refining existing skills or extending with new capabilities. The resulting skills are maintained in a shared repository and synchronized across users, enabling cross-user knowledge transfer and cumulative capability improvement without additional user effort. Experiments on WildClawBench show that with limited interaction and feedback, SkillClaw significantly improves the performance of Qwen3-Max in real-world agent scenarios.

Key Features

Continuous aggregation of cross-user trajectories

Autonomous evolver that identifies recurring behavioral patterns

Refines existing skills or extends them with new capabilities

Shared skill repository synchronized across all users

Cross-user knowledge transfer and cumulative capability improvement

Works with limited interaction and feedback to improve performance

Pros & Cons

Pros

Automatically aggregates user interactions to improve skills without manual curation
Enables system-wide propagation of improvements discovered in one context
Works with limited feedback to significantly boost performance
Reduces redundant rediscovery of workflows, tool usage, and failure modes across users

Cons

Framework is a work in progress and may require additional engineering for production deployment
Performance gains demonstrated only on specific benchmark (WildClawBench) and model (Qwen3-Max)
Relies on multi-user interaction data; single-user scenarios may not benefit as much

Best For

Improving LLM agent performance in real-world tasksEnabling cross-user knowledge transfer in multi-agent systemsAutomated skill refinement from user interactions over timeScaling agent capabilities without manual curation

FAQ

How does SkillClaw evolve skills?

It continuously aggregates trajectories from user interactions, and an autonomous evolver identifies recurring behavioral patterns, refining existing skills or extending them with new capabilities.

Does SkillClaw require extra effort from users?

No, improvements propagate system-wide without requiring additional effort from users. The system automatically learns from user interactions.

What model was used in experiments?

The experiments used Qwen3-Max on the WildClawBench benchmark, demonstrating significant performance improvements with limited interaction and feedback.