autoresearchclaw-metaclaw-cross-run-skill-evolution
AutoResearchClaw implements MetaClaw — a cross-run skill evolution store with time-decay weighting that compounds lessons across separate research runs. This differs from Oracle’s Vault + Graphiti dual-write in that MetaClaw specifically tracks per-experiment strategy effectiveness and applies decay to older lessons. The Oracle could adopt this time-decay pattern in Graphiti for the AutoResearch experimentation domain.