OpenSkill
NewOpen-World Self-Evolution for LLM Agents — agents that build both their skills and their own verification signals from scratch, with no target-task supervision. (Code coming soon.)
Summary
OpenSkill enables LLM agents to autonomously evolve their capabilities and generate their own verification signals without any task-specific supervision.
- This skill is useful for developers who want to build self-improving agents that can adapt to new challenges in open-world environments.
Install & Usage
mkdir -p .claude/skillsmkdir -p .claude/skills && curl -o .claude/skills/openskill.md https://raw.githubusercontent.com/OpenLAIR/OpenSkill/main/SKILL.md/openskillUse Cases
Usage Examples
/openskill evolve --domain minecraft --steps 1000
Initialize an OpenSkill agent in a grid world and let it autonomously discover navigation skills.
/openskill verify --task "open chest" --checklist "approach, interact, confirm"
Security Audits
Frequently Asked Questions
What is OpenSkill?
OpenSkill enables LLM agents to autonomously evolve their capabilities and generate their own verification signals without any task-specific supervision. This skill is useful for developers who want to build self-improving agents that can adapt to new challenges in open-world environments.
How to install OpenSkill?
To install OpenSkill: create the skills directory (mkdir -p .claude/skills), then run: mkdir -p .claude/skills && curl -o .claude/skills/openskill.md https://raw.githubusercontent.com/OpenLAIR/OpenSkill/main/SKILL.md. Finally, /openskill in Claude Code.
What is OpenSkill best for?
OpenSkill is a skill categorized under General. It is designed for: agent. Created by OpenLAIR.
What can I use OpenSkill for?
OpenSkill is useful for: Automatically generate new skills for an agent based on observed gaps in its performance during exploration tasks.; Create self-supervised verification signals to validate agent behavior without human-labeled data.; Enable an agent to iteratively refine its own skill set through trial and error in a simulated open world.; Develop agents that can discover and learn novel tasks without predefined reward functions.; Build autonomous systems that adapt to dynamic environments by evolving their capabilities over time.; Implement a curriculum learning approach where the agent generates its own training tasks and checks its progress..