AgentSkillsCN

run

端到端运行Playwright UI测试，收集测试产物，并总结失败情况，提出下一步的修复方案。

SKILL.md
--- frontmatter
name: run
description: Run Playwright UI tests end-to-end, collect artifacts, and summarize failures with next-step fixes.
disable-model-invocation: true
argument-hint: "[optional: test-path or grep] [--headed] [--project=...]"
allowed-tools: Read, Grep, Glob, Bash(npm *), Bash(pnpm *), Bash(yarn *), Bash(npx playwright *), Bash(node *)

Goal

Run Playwright tests deterministically and return a pass/fail summary, the top failing tests with likely cause, and artifact paths (playwright-report/, test-results/, traces).

Rules

  • •Prefer npm scripts if present in package.json, else use npx playwright.
  • •Default to headless unless user passed --headed.
  • •Only run against localhost / internal env configured in playwright.config.*.
  • •If Playwright or browsers are missing, install them using scripts/ensure-playwright.sh.
  • •Always run with trace enabled on failure if not already configured.

Steps

1. Locate Playwright

  • •If package.json exists and includes @playwright/test, use project-local.
  • •Otherwise install as a dev dependency + browsers.

2. Ensure browsers installed

bash
bash plugins/playwright/skills/run/scripts/ensure-playwright.sh

3. Run tests

  • •If package.json has a test:ui script, run npm run test:ui -- $ARGUMENTS
  • •Otherwise run npx playwright test $ARGUMENTS

4. If failing — re-run with diagnostics

Re-run only failing tests once:

bash
npx playwright test --last-failed --trace=on --screenshot=only-on-failure

5. Summarize

  • •Give a tight summary and list failing spec names.
  • •Point to artifacts: playwright-report/ and test-results/ paths.
  • •Follow the structured output contract in references/output-format.md.

Constraints

  • •Never open arbitrary external URLs.
  • •Do not start application servers unless explicitly instructed.
  • •Keep trace files for debugging; do not delete them automatically.