← ClaudeAtlas

loadlisted

Design and run performance tests — smoke, load, stress, spike, soak, breakpoint — with k6 by default, from a workload model based on the capacity model, with SLO thresholds, open-model executors, server-side monitoring, honest environment caveats and CI wiring. Use for "load test the API", "stress test", "will it handle N users", "soak test".
lucasvmigotto/ai-gent · ★ 0 · AI & Automation · score 63
Install: claude install-skill lucasvmigotto/ai-gent
# Load and performance tests ## Before starting 1. Read `../../references/qa-quality.md` and `../../references/pipeline-digest.md` (the short form; the full `../../references/pipeline.md` only when the digest conflicts with or is silent on your case). 2. Read: `docs/product/architecture.md` (capacity model, SLOs, evolution triggers), the brief's NFRs, `specs/*/qa.md` load profiles, `contracts/openapi.yaml`, `backend.md` performance budgets, and `delivery.md` (environments). 3. **Authorization and target.** Only test systems the user owns and approves, never production or third-party APIs without written approval; stub external dependencies (WireMock/Prism from `devcontainer:infra`) so the test measures this system, not a vendor. ## Workload model Research the stack's benchmarks, k6's current version and executor semantics, and the vendor's published rate limits (`../../references/research.md`, about 5 searches and 5 fetches). Never exceed a published limit and never research how to bypass a limiter; authorization stays in the step above. Query tool and vendor names only. Cite source and date in the report and the README, labelled `[RESEARCHED: S<n>]`, `[KNOWN]` or `[ASSUMPTION]`; with no web tools, a limit claim is `[ASSUMPTION]`, "verify before relying". Before any script, write the model in `qa.md` / a `tests/load/README.md` (conventions in `../../references/readme.md`; never run against production is a `[!CAUTION]` there): - **Arrival rate** (reques