菜单
Set the performance strategy before testing: turn vague goals into falsifiable NFRs (percentile + threshold + load), model the real workload, link targets to the customer SLO, and choose the test type (load/stress/soak/spike) that answers the open question.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/mcorbett51090/RavenClaude --skill performance-test-strategy命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Route everyday, well-defined work (single-file edits, test-writing, summarization, mechanical refactors, boilerplate) to a cheaper coding agent (Grok or Copilot CLI) instead of doing it in the main Claude session, escalating multi-file/architecture/debugging/security work to Claude. Opt-in via the `cheap_lane:` comfort-posture knob.
Authors, validates and packages a claude.ai Organization Skill — the intake questions, the trigger and scope exercises, then lint, pack and verify against measured platform constraints. Reach for this when someone wants to publish a Skill to their organization, or when an upload was rejected and the reason is unclear.
Drafts a reply to an inbound customer email, grounded in the account record and the published help-centre articles rather than in recall. Use when a customer emails about billing, a renewal or an outage, when a reply must cite current policy, or when a teammate asks for a first draft to edit.
基于 SOC 职业分类
正在显示 SKILL.md
| name | performance-test-strategy |
| description | Set the performance strategy before testing: turn vague goals into falsifiable NFRs (percentile + threshold + load), model the real workload, link targets to the customer SLO, and choose the test type (load/stress/soak/spike) that answers the open question. |
Every NFR is a percentile + a threshold + the load it holds at — "p99 ≤ 200 ms at 5,000 req/s, 70/30 read/write", never "the API should be fast". A target with no load attached can't be passed, failed, or sized from. Set targets on p95/p99 (and max), never the average.
The traffic mix, arrival pattern, data distribution, cache warmth, and peak multiplier drive the result. Specify them first: request weights, open vs. closed arrival, steady vs. peak, data skew/cardinality, and the cache-warmth assumption. The workload model is the contract you hand to the load-testing-engineer.
Your NFRs are the engineering targets that keep the customer SLO/error budget safe. Make the two consistent; the SLO itself is owned by observability-sre. Name which is latency-protected and which is throughput-protected — you can't max both.
Load proves the steady-state target; stress finds the knee; soak finds the slow leak/degradation; spike proves elasticity. Name the unanswered question and pick the test that answers it — one steady-state load run is not a performance sign-off.
A performance test plan: the NFR targets, the workload model, the test type(s) and sequence, the environment + data plan, and the pass/fail thresholds. Hand the test build to load-testing-engineer; route the SLO to observability-sre.