WildClawBench: A Real-World, Long-Horizon Benchmark for AI Agents

New benchmark and dataset for robotic manipulation in unconstrained 'wild' environments. Includes standardized containers, leaderboards, and evaluation protocols for cross-embodiment policies.