Benchmark harness for A/B testing Claude Code plugins against OOLONG long-context reasoning tasks. Compare truncation vs RLM-RS recursive chunking strategies. Features Claude Code hooks integration, SQLite persistence, and comprehensive scoring aligned with the OOLONG paper methodology.