{"slug":"jeremylongshore-coreweave-fabric-diagnostics","source_name":"jeremylongshore/coreweave-fabric-diagnostics","name":"Jeremylongshore/Coreweave Fabric Diagnostics","description":"Diagnose the most expensive silent failure on a CoreWeave multi-node GPU job:\nGPUDirect RDMA falling back from InfiniBand to TCP. When NCCL drops from NET/IB to\nNET/Socket, collectives keep running with NO error but throughput collapses (commonly\n5-20x slower) while every GPU still bills at full rate — 5x the GPU bill for the same\nwork, invisibly. Paste an NCCL_DEBUG=INFO log (and/or a pod-spec, ibstat, or\nall_reduce_perf output) and the bundled deterministic script verdicts whether RDMA is\nactu","version":1,"lift":{"pass_rate_delta_pts":45.45,"pass_rate_pct":90.9,"total_cases":22,"passed_cases":20,"tokens_delta_pct":155.5,"turns_delta_pct":0,"verdict":"mixed","benchmark_model":"gemini-3.6-flash","grading_method":"judged","completed_at":"2026-08-14T11:44:09.310198+00:00"},"skill_score":0.9091,"benchmark_models":[{"model":"gemini-3.6-flash","headline":true,"delta_pts":45.45,"with_pass_pct":90.9,"without_pass_pct":45.5,"tokens_delta_pct":155.5,"turns_delta_pct":0,"total_cases":22,"cases_aggregated":21,"verdict":"mixed","never_hurt":true,"completed_at":"2026-08-14T11:44:09.310198+00:00","run_id":"803dcb05-9c7a-484c-bf99-40eac0069562","version_number":1,"is_latest_version":true,"gate":null}],"trust":{"skill_safety":"passed","safety_status":"clean","intent_verdict":"safe","content_status":"clean","indexable":true},"license":"MIT","install_count":0,"manifest_hash":"06538aa42535cf2f22780b1a0dba6b49bdcf60e2ec681b7811b8f24d138c983f","raw_url":"https://app.decimal.ai/s/jeremylongshore-coreweave-fabric-diagnostics/SKILL.md","scorecard_url":"https://app.decimal.ai/skills/jeremylongshore-coreweave-fabric-diagnostics"}