What problem does it solve? Running credible production benchmarks against the GUARDIAN server requires coordinating ephemeral ECS/Fargate workers, selecting the right signer and workload profiles, cleaning up benchmark-owned data, and interpreting throughput and latency results correctly. This Skill encodes that entire workflow so benchmark runs are reproducible, safe, and consistently reported. ## Core Features & Use Cases - Guided benchmark execution: Walks through readiness checks, local preflight, profile selection, and the canonical run-prod-benchmark-ecs.sh execution path for ECDSA, Falcon, and mixed gRPC workloads. - Cleanup and artifact validation: Verifies cleanup-manifest.json completion, run reports, and endpoint health after each run so no benchmark-owned data or temporary AWS resources linger. - Result interpretation and reporting: Applies documented rules for throughput, latency percentiles, state_conflict caveats, shard loss, and 500 TPS sizing estimates, following the established report structure. - Use Case: An engineer needs to know how many GUARDIAN tasks are required for a 500 TPS target. The Skill selects the Falcon+ECDSA mixed profile, runs 16 ECS workers against prod, validates cleanup, and produces a report with throughput, p50/p95/p99 latency, and a sizing estimate. ## Quick Start Use the run-guardian-prod-benchmarks skill to execute the falcon-ecdsa-mixed-burst-scale profile against the prod GUARDIAN endpoint and summarize the throughput and latency results.
