DD-18 — MetaClaw: The Self-Evolving Proxy Harness

MetaClaw: The Self-Evolving Proxy Harness

3,500+ stars. Python/FastAPI. UNC AIMING Lab. The proxy that meta-learns from every conversation via skill injection + cloud LoRA/RL.

60
minutes
3
artifacts
Rubric Score

29/60 · Highest Module 4 Memory (4/5) · Learning proxy NOT governance proxy — dual-use risk surface

The Score
Architect's Verdict

makes any agent self-improving via proxy-layer skills + idle-time RL; same position as governance proxies but opposite purpose

Artifacts