菜单
Write and run Artillery load tests with YAML phases and scenarios, CSV data payloads, the expect plugin for functional checks, and ensure thresholds that fail CI when latency or error budgets are breached.
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/PramodDutta/qaskills --skill load-test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。
基于 SOC 职业分类
正在显示 SKILL.md
| name | Artillery Load Testing |
| description | Write and run Artillery load tests with YAML phases and scenarios, CSV data payloads, the expect plugin for functional checks, and ensure thresholds that fail CI when latency or error budgets are breached. |
| version | 1.0.0 |
| author | thetestingacademy |
| license | MIT |
| tags | ["artillery","load-testing","performance","stress-testing","yaml","thresholds","ci","api","latency"] |
| testingTypes | ["load","performance"] |
| frameworks | ["k6"] |
| languages | ["javascript"] |
| domains | ["api","web"] |
| agents | ["claude-code","cursor","github-copilot","windsurf","codex","aider","continue","cline","zed","bolt","gemini-cli","amp"] |
This skill makes an AI agent author Artillery 2.x load tests as declarative YAML: realistic traffic phases, multi-step scenarios with captured variables, CSV-driven virtual user data, functional expect checks inside load flows, and ensure thresholds that turn latency regressions into red CI builds. Trigger it when a project contains artillery.yml, artillery in package.json, or when the user asks for load, stress, soak, or spike testing of an HTTP API or website in a Node.js stack.
arrivalRate hides cold-start effects and autoscaling lag. Always define at least warm-up, ramp, and sustain phases.arrivalRate is new virtual users per second, not concurrency. Each arriving VU runs the whole scenario. If your scenario takes 10 seconds and you arrive 50/sec, you have roughly 500 concurrent users. Calculate this before picking numbers.ensure plugin so artillery run exits non-zero when p95/p99 or error rate budgets are blown. A load test that cannot fail is a demo, not a test.expect. A server returning 200 with an empty body at p99 latency is still broken. Check statusCode, contentType, and hasProperty inside the flow.payload with hundreds of distinct credentials and SKUs to defeat caches realistically.npm install --save-dev artillery@latest
npx artillery version
# Run a test locally
npx artillery run load/checkout.yml
# Save raw metrics for trend analysis
npx artillery run load/checkout.yml --output artillery-report.json
# load/checkout.yml
{}
{}
email,password
loadtest-001@example.com,Str0ngPass!001
loadtest-002@example.com,Str0ngPass!002
loadtest-003@example.com,Str0ngPass!003
loadtest-004@example.com,Str0ngPass!004
Generate hundreds of rows with a one-liner instead of writing them by hand:
seq -w 1 500 | awk -F, 'BEGIN{print "email,password"} {printf "loadtest-%s@example.com,Str0ngPass!%s\n", $1, $1}' > users.csv
# In config:
config:
target: https://staging-api.example.com
processor: ./processor.js
scenarios:
- name: create-order-with-dynamic-payload
flow:
- function: generateOrderPayload
- post:
url: /orders
json:
sku: '{{ sku }}'
quantity: '{{ quantity }}'
afterResponse: logSlowResponse
// processor.js
module.exports = { generateOrderPayload, logSlowResponse };
function generateOrderPayload(context, events, done) {
// Runs before the request; sets template variables on the VU context
context.vars.sku = `SKU-${1000 + Math.floor(Math.random() * 9000)}`;
context.vars.quantity = 1 + Math.floor(Math.random() * 4);
return done();
}
function logSlowResponse(requestParams, response, context, events, done) {
const tookMs = response.timings ? response.timings.phases.total : 0;
if (tookMs > 1000) {
console.warn(`SLOW ${requestParams.url} -> ${response.statusCode} in ${tookMs}ms`);
events.emit('counter', 'custom.slow_responses', 1);
}
return done();
}
The ensure plugin makes Artillery exit with code 1 on any threshold breach, so the job fails without extra scripting.
# .github/workflows/load-test.yml
name: load-test
on:
workflow_dispatch:
schedule:
- cron: '0 2 * * 1'
jobs:
artillery:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with:
node-version: 20
cache: npm
- run: npm ci
- name: Run load test against staging
run: npx artillery run load/checkout.yml --output artillery-report.json
env:
ARTILLERY_DISABLE_TELEMETRY: 'true'
- name: Upload raw metrics
if: always()
uses: actions/upload-artifact@v4
with:
name: artillery-report
path: artillery-report.json
think steps (1-3 seconds) between requests so VUs pace like humans instead of a retry storm.weight: to mirror real traffic mix (for example 70 percent browse, 25 percent search, 5 percent checkout).--overrides '{"config":{"phases":[{"duration":10,"arrivalRate":1}]}}') before any big run to catch broken auth and 4xx noise cheaply.--output JSON artifacts from every CI run so you can diff p95 across releases.http.timeout explicitly; the 120-second default hides hangs as slow successes.arrivalRate and no ramp: you are testing the load balancer's SYN queue, not your application.expect or ensure block exists.arrivalRate past what one generator machine can produce: watch for Artillery's own CPU warnings, and split load across workers (or Fargate via artillery run-fargate) instead.artillery.yml, files under load/ or perf/ with Artillery config, or artillery in package.json devDependencies.