Query Seeks Benchmarks for Agentic Debugging