메뉴
Fairness + reproducibility review for exeris-benchmarks. Use on every committed result and every cross-target claim — enforces matched scoping, separation axes, and complete metadata.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
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.
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.
Guard Exeris cross-tier comparisons so Community vs Enterprise claims stay correctly labeled, protocol-consistent, and inference-safe.
| name | exeris-benchmarks-fairness-reproducibility-review |
| description | Fairness + reproducibility review for exeris-benchmarks. Use on every committed result and every cross-target claim — enforces matched scoping, separation axes, and complete metadata. |
Enforce: matched payload / concurrency / protocol / target for cross-target claims; mandatory separation axes always labelled; every committed result captures complete reproducibility metadata.
results/raw/, results/reports/, results/history/, results/constrained/, results/verify-phasemarker/, results/quick-verify-fix-jfr/.baselines/<repo>/<mode>/<hardware>.json.schemas/ (benchmark-result.schema.json, benchmark-env.schema.json, comparative-result.schema.json, fairness-index.schema.json, runtime-execution-profile-matrix.schema.json, enterprise-target-contract.schema.json).-f 3 minimum for publishable runs. -f 1 in committed results → reject. Standard JVM flags present.docs/regression-policy.md; baseline never silently refreshed to mask regression.APPROVE / CONDITIONAL / REJECT.-f 3+; baseline cite if applicable.-f 1 published; silent baseline update.-f value)APPROVE / CONDITIONAL / REJECT)-f 1 in a published run.