OpenSpace
Self-evolving skill engine for AI agents by HKUDS. Instead of stacking frozen skills, OpenSpace loops on evolution: broken skills self-repair, enhanced versions derive themselves, novel patterns get captured. Cloud community registry to share evolved skills. Python backend + LiteLLM, React dashboard, SQLite persistence, MCP integration with Claude Code, Cursor, nanobot.
Strengths
- Self-evolving skills: repair, derivation, novel pattern capture without retraining
- MCP-compatible, plugs into existing runtimes (Claude Code, Cursor, nanobot)
- Cloud community registry: upload/download evolved skills across teams
- Measured 46% token reduction via mature skill reuse, quantified ROI
Limitations
- Requires separate LLM credentials (LiteLLM), not cost-autonomous
- Heavy stack: Python backend + React dashboard + SQLite, beyond a simple .claude/
- Young project (7.1k stars, early 2026), maturity and registry longevity to confirm
- Learning curve to define skills and configure evolution rules
Best for
- Teams running AI agents in production wanting continuous improvement without manual retraining
- Organizations prioritizing token efficiency and measurable ROI on AI automation
- Developers seeking a collaborative skill ecosystem rather than siloed capabilities