AGENT SKILL

playwright-verify

Browser verification using Playwright MCP. Navigate, interact, screenshot.

License 许可
MIT
所属仓库
vyzn-tech/po-agent
最近更新
2026-03-11
Stars
★ 0
Forks
1

01 前言 Front Matter

name
playwright-verify
description
Browser verification using Playwright MCP. Navigate, interact, screenshot.
allowed-tools
Bash

02 资源文件 13

03 SKILL.md 正文

Playwright Verification

Available Tools

Tool Purpose Use when
browser_navigate Navigate to a URL Starting verification
browser_snapshot Accessibility tree (~1-2 KB text) Understanding page state — preferred
browser_click Click elements Interacting with UI
browser_type Type into fields Filling forms
browser_take_screenshot Capture page as image (~1-2 MB) Visual confirmation — sparingly
browser_run_code Execute JavaScript Saving screenshots to disk

Best Practices

Prefer browser_snapshot over browser_take_screenshot

  • browser_snapshot returns text (~1-2 KB) — use for understanding state
  • browser_take_screenshot returns an image (~1-2 MB) — use only for visual confirmation

Limit inline screenshots to ~3 per session

Exceeding this risks 413 payload errors that terminate the session.

Save bulk evidence to disk

// browser_run_code:
await page.screenshot({ path: '/tmp/agent-artifacts/screenshots/step-name.png', fullPage: true });

Save to /tmp/agent-artifacts/screenshots/ — the workflow auto-uploads these.

Verification Flow

  1. Navigate to the target URL
  2. Use browser_snapshot to understand page state
  3. Interact and verify the change works
  4. Save evidence screenshots to disk
  5. Use browser_take_screenshot for the ONE key screenshot (max ~3 inline)

04 同仓库更多 Skills