メニュー
Review Exeris benchmark methodology for fairness and evidence quality: warmup, measurement windows, target/payload/concurrency equivalence, and overclaiming risk.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | exeris-bench-methodology-review |
| description | Review Exeris benchmark methodology for fairness and evidence quality: warmup, measurement windows, target/payload/concurrency equivalence, and overclaiming risk. |
| argument-hint | Provide scenario, compared targets/protocols, run config, and intended conclusion. |
SOUND, CONDITIONALLY_SOUND, or UNSOUND.Comparative strict-gate review for exeris-benchmarks. Use on every PR making a cross-target comparative claim — verifies 4 gate artefacts present + `claim-status.json = comparison_eligible` + `track_id` consistent.
Fairness + reproducibility review for exeris-benchmarks. Use on every committed result and every cross-target claim — enforces matched scoping, separation axes, and complete metadata.
Publication-mode scrub review for exeris-benchmarks. Use whenever `publish-report.sh` runs, when an artefact crosses to a publishable surface, or when raw `.jfr` / flamegraph appears.
Router/Planner skill for exeris-benchmarks. Produces primary agent, secondary handoffs, execution order, validation gates, and minimal next action for a benchmark task.
Router/Planner triage skill for exeris-benchmarks. Classifies task type (architecture / scenario / driver / JMH / target / verification / docs / publication), scope, severity, and recommends primary agent based on primary risk against fairness / reproducibility / strict gate / confidentiality.
TLS comparator label discipline review for exeris-benchmarks. Use on every TLS JMH benchmark / TLS row / TLS report change.