leaklens

The two tools split cleanly along aggregation level: benchmark_leakage_summary reports benchmark-wide rates and optional per-repo flagged task lists, while check_task_leakage returns per-task evidence for 1-50 IDs. There is a small overlap where both can surface flagged task identities, but granularity and use-case differ enough that misselection is unlikely.

Both names are snake_case and share the 'leakage' domain noun, making them visually related. The minor deviation is structural: check_task_leakage is verb-first while benchmark_leakage_summary is noun-first, so the pattern is not strictly uniform.

Two tools is thin in absolute terms, but the server's scope is deliberately narrow (leakage measurement over SWE-bench-style benchmarks), so each tool carries real weight. Still, two endpoints is borderline sparse and forces bulk work through a single per-task tool.

The core lifecycle (benchmark-level summary, task-level check, bulk up to 50 IDs, ready-to-cite output) is covered, but there is no way to resolve or revisit disputed flags and the referenced auditYourOwn capability is explicitly absent from the surface. Agents can work around this but hit a dead end on self-audit and dispute handling.