菜单
Automated browser testing and debugging using Browser DevTools MCP. Use when testing web pages, debugging frontend issues, verifying UI behavior, or automating browser interactions.
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。
基于 SOC 职业分类
Use when the user wants to run JavaScript in the browser, run code in Node, or batch multiple MCP tool calls. Run script, execute code, batch tools.
Debug Node.js backend processes using non-blocking tracepoints, logpoints, exceptionpoints, and JavaScript execution. Use when debugging API servers, workers, or backend scripts.
Debug web applications by inspecting console logs, network requests, JavaScript errors, and using non-blocking tracepoints/logpoints. Use when debugging web pages, inspecting API calls, or tracing code execution.
Monitor and debug web applications using OpenTelemetry, console logs, network requests, and distributed tracing. Use when correlating frontend and backend requests or debugging request flow across services.
Analyze web page performance using Web Vitals and network timing metrics. Use when optimizing load times, checking Core Web Vitals, or investigating slow pages.
Perform visual testing and UI verification using screenshots and DOM inspection. Use when verifying UI appearance, comparing with designs, or checking for visual regressions.
| name | browser-testing |
| description | Automated browser testing and debugging using Browser DevTools MCP. Use when testing web pages, debugging frontend issues, verifying UI behavior, or automating browser interactions. |
Automated browser testing and debugging skill using Browser DevTools MCP.
Page structure: To understand what is on a page, use ARIA snapshot first (a11y_take-aria-snapshot or a11y_take-ax-tree-snapshot), not screenshot. Use screenshot only when you need to verify how something looks visually (e.g. design check, visual bug).
This skill activates when:
a11y_take-aria-snapshot) — Returns refs (e1, e2, …) and a refs map. Options: interactiveOnly, cursorInteractive (for div/span buttons), selector (scope).e1, @e1). Refs valid until next snapshot or navigation.content_take-screenshot with annotate: true overlays labels [1],[2],… mapping to e1,e2,…navigation_go-to) — returns ARIA snapshot and refs by default; use response refs for interactions without calling a11y_take-aria-snapshot againnavigation_go-back-or-forward and navigation_reload (includeSnapshot: true)e1, #submit)interaction_press-key), scroll (interaction_scroll), drag (interaction_drag)content_take-screenshot) — optional annotate, annotateContent, cursorInteractiveo11y_get-console-messages)o11y_get-http-requests)o11y_get-web-vitals)o11y_get-trace-id, o11y_set-trace-id)a11y_take-aria-snapshot) — returns refs; options: interactiveOnly, cursorInteractive, selectora11y_take-ax-tree-snapshot) — bounding box, occlusion, stylesstub_intercept-http-request)stub_mock-http-response)stub_list)stub_clear)react_get-component-for-element)react_get-element-for-component)page.evaluate() (e.g. return await page.evaluate(() => document.title)). See commands/browser/execute.md for callTool and page.page; use built-ins and callTool). On Node platform, execute runs in the session VM (callTool only).execute — Run custom JavaScript that can call multiple tools in one request via await callTool(name, input, returnOutput?). Use to batch click/fill/select/navigation and reduce round-trips. Code runs in async IIFE (no wrapper); browser platform exposes page (Playwright). Prefer when you have a fixed sequence of steps; use individual tools when you need to inspect results between steps. Also supported: browser-devtools-mcp CLI subcommand run execute, and daemon HTTP API (POST /call with toolName: "execute").a11y_take-aria-snapshot, then use refs (e1, e2) in interactions or take annotated screenshots.annotate: true when you need visual labels that match refs for vision-driven actions