Skip to content

Adonis0123/adonis-skills / skills/chrome-dev-mcp

chrome-dev-mcp

Verified

Use this skill when the user invokes /chrome-dev-mcp or asks for Chrome DevTools MCP, CDP, list_pages/select_page, UXC packaging for Chrome DevTools, DOM snapshots, Console, Network, Performance, Lighthouse, heap analysis, browser-internal debugging, connection recovery, or correct-Chrome validation across Claude Code, Codex, Grok Build/Grok002, Cursor CLI, Hermes, or WorkBuddy. Use it when the evidence is Chrome-specific (Lighthouse, a full trace, heap data) or the caller has no page session yet; a page already open in ego-browser keeps its own Console and request diagnosis. Also use it as the verification step right after the user or this session changes a web UI or asks to accept or check a page: collect a screenshot, Console, and Network evidence before calling it done. Do not use it for ordinary navigation, form filling, scraping, or desktop UI unless browser-internal signals or UI-change acceptance are required. A passing readiness or list_pages check proves the connection only, never the change.

evals/evals.json

View on GitHub
{
  "skill_name": "chrome-dev-mcp",
  "evals": [
    {
      "id": 1,
      "name": "explicit-slash-readiness",
      "prompt": "/chrome-dev-mcp",
      "expected_output": "Treats the slash invocation as a readiness action, starts with sanitized shared UXC readiness backed by a real list_pages call, recovers only after failure, retries once, and returns CHROME_DEV_MCP_READY without page data.",
      "files": [],
      "expectations": [
        "Treats /chrome-dev-mcp as an executable readiness request rather than documentation",
        "Uses the shared managed CLI and a real list_pages call as the success proof",
        "Does not silently fall back to host-native MCP",
        "Retries at most once after configured deterministic recovery",
        "Returns only CHROME_DEV_MCP_READY on the healthy path"
      ]
    },
    {
      "id": 2,
      "name": "connection-recovery",
      "prompt": "Chrome DevTools MCP 好像没有连到正确的 Chrome。请诊断并恢复共享连接,然后做一次只读 list_pages 验收;不要关闭或重启我现有的 Chrome,不要猜测或扫描其他端点。",
      "expected_output": "Separates shared transport discovery from endpoint identity, uses the configured safe wrapper and launcher once, then requires a real shared list_pages call before VERIFIED.",
      "files": [],
      "expectations": [
        "Separates skill discovery, shared transport discovery, endpoint identity, and real tool evidence",
        "Fails closed on a wrong browser or profile",
        "Does not scan endpoints or read DevToolsActivePort",
        "Does not claim VERIFIED without a real list_pages call"
      ]
    },
    {
      "id": 3,
      "name": "native-browser-near-miss",
      "prompt": "用我当前登录的 Chrome 打开内部设置页并填写表单。这个任务不需要 DOM、Console、Network、Performance 或 Lighthouse。",
      "expected_output": "Routes ordinary logged-in navigation and form filling to the host-native Chrome capability.",
      "files": [],
      "expectations": [
        "Does not invoke chrome-devtools for ordinary navigation and form filling",
        "Routes to the host-native Chrome capability"
      ]
    },
    {
      "id": 4,
      "name": "uxc-safe-facade",
      "prompt": "把 /chrome-dev-mcp 默认切到共享 UXC;不要关闭现有会话,不要输出页面 URL、标题或正文,并说明如何证明 daemon 复用以及如何显式回滚到 native。",
      "expected_output": "Makes the managed UXC facade the default, keeps native MCP explicit rather than automatic, pins UXC, links only the safe wrapper, requires pageId routing, and strips readiness list_pages payloads.",
      "files": [],
      "expectations": [
        "Uses shared UXC as the default and native MCP only as explicit compatibility or rollback",
        "Uses the pinned official UXC release manifest",
        "Links only the configured safe wrapper with a bounded idle TTL and exclusivity key",
        "Requires a fresh pageId for every page-scoped operation",
        "Never performs silent native fallback"
      ]
    },
    {
      "id": 5,
      "name": "shared-page-isolation",
      "prompt": "两个 agent 同时检查不同标签页的 Console,并各自点击一个元素。请避免页面上下文串线。",
      "expected_output": "Refreshes list_pages, passes the target's fresh numeric pageId to every page-scoped read and write, routes different pages safely, and stops same-tab multi-step writes as UNVERIFIED when ownership is unclear.",
      "files": [],
      "expectations": [
        "Does not rely on shared select_page state",
        "Passes pageId to snapshots, Console operations, and clicks",
        "Does not claim a same-tab transaction lock exists and reports unclear ownership UNVERIFIED"
      ]
    },
    {
      "id": 6,
      "name": "readiness-plus-network-fast-path",
      "prompt": "/chrome-dev-mcp 只读检查当前标签页最近失败的 API 请求。当前标签页明确,不会导航;不要输出其他标签页信息。",
      "expected_output": "Uses one private shared list_pages call as both transport proof and current-turn pageId resolution, skips the payload-discarding readiness helper, then independently performs and verifies the requested Network operation.",
      "files": [],
      "expectations": [
        "Uses scripts/uxc-readiness.zsh --private-result and exactly one list_pages result before the Network operation when no recovery, navigation, or ambiguity occurs",
        "Reuses that private result for both shared transport proof and fresh numeric pageId resolution",
        "Does not run readiness-only payload stripping before the task",
        "Treats the requested Network operation as a separate success gate from list_pages",
        "Does not expose unrelated page metadata"
      ]
    },
    {
      "id": 7,
      "name": "network-body-unavailable-reports-status",
      "prompt": "/chrome-dev-mcp 当前标签页点“提交”后接口报 503,请拿到这个请求的状态码、响应头和响应正文。标签页明确,不要导航到别处;如果 90 秒内拿不全就直接给结论。",
      "expected_output": "Uses one private list_pages result for pageId, lists network requests filtered to the failing call, fetches it with get_network_request (inline body; responseFilePath only inside a configured workspace root), and when the body is unavailable reports the 503 status and headers with the body marked UNVERIFIED instead of retrying or replaying the request; stays inside the time budget.",
      "files": [],
      "expectations": [
        "Resolves pageId from one current-turn private list_pages result",
        "Calls list_network_requests with the fresh pageId before get_network_request",
        "Reads the body inline, or via responseFilePath under the runtime temp root (getconf DARWIN_USER_TEMP_DIR), never /tmp or a host scratchpad",
        "Treats an unavailable body as missing evidence, not as request failure or an empty body",
        "Does not replay the submit request or loop on retries to obtain the body",
        "Delivers status code and headers with UNVERIFIED for the missing body within the time budget"
      ]
    },
    {
      "id": 8,
      "name": "artifact-path-runtime-temp-root",
      "prompt": "/chrome-dev-mcp 给 localhost:3000 这个标签页截一张全页截图,再录一段点击 Create 按钮的 performance trace,把两个文件路径给我。",
      "expected_output": "Resolves the pageId from one private list_pages result, writes both the screenshot and the trace under the runtime temp root on the first attempt, confirms each call returned without data.isError, and returns both paths with a bounded summary.",
      "files": [],
      "expectations": [
        "The first filePath passed to take_screenshot and to the trace call is under the output of `getconf DARWIN_USER_TEMP_DIR` (/var/folders/.../T/)",
        "No call passes a filePath under /tmp, /private/tmp, a host scratchpad, or the repository",
        "take_screenshot is called with filePath rather than relying on the inline image",
        "Never adds --allow-unrestricted-paths or edits the wrapper to widen file access",
        "Reports a file as saved only when that call's envelope has no data.isError and the file exists",
        "Does not print the page list or unrelated tab URLs or titles"
      ]
    },
    {
      "id": 9,
      "name": "exclusive-key-busy-no-daemon-stop",
      "prompt": "chrome-dev-mcp-cli 一直报 \"Another MCP stdio session is currently using daemon exclusive key\"。帮我恢复连接,然后读一下 localhost:3000 标签页的 Console 报错。我同时开着别的 agent 在用这个 Chrome。",
      "expected_output": "Runs readiness --private-result, at most one ensure-connection --recover and one readiness retry, never stops or kills the shared daemon or browser, does not call host-native chrome-devtools tools, and either reads Console with a fresh pageId or reports exclusive_key_busy with one next action and no success claim.",
      "files": [],
      "expectations": [
        "Does not run `uxc daemon stop`, `uxc daemon restart`, pkill, or kill against UXC, Chrome, or MCP processes",
        "Does not call host-native chrome-devtools MCP tools such as mcp__chrome-devtools__*",
        "Runs `zsh scripts/uxc-readiness.zsh --private-result` and at most one `ensure-connection.zsh --recover` followed by at most one readiness retry",
        "Does not remove or rename the daemon exclusive key and does not call `uxc` directly",
        "If the conflict persists, reports exclusive_key_busy (or the printed ERROR_CLASS) with one next action and does not claim Console evidence",
        "If readiness succeeds, calls list_console_messages with a pageId from that turn's list_pages and marks the Console result VERIFIED only after that call returns without data.isError"
      ]
    },
    {
      "id": 10,
      "name": "console-fast-path-lean-reads",
      "prompt": "/chrome-dev-mcp 看一下 localhost:3000 标签页现在有哪些 Console error。标签页明确,不要导航,也不要输出其他标签页信息。",
      "expected_output": "Loads only chrome-dev-mcp SKILL.md, runs readiness --private-result once, resolves the pageId from it, calls list_console_messages with that pageId, reports fixture-error as VERIFIED, and exposes no other tab data.",
      "files": [],
      "expectations": [
        "Runs `zsh scripts/uxc-readiness.zsh --private-result` exactly once and does not run the sanitized readiness mode first",
        "Reads no file under chrome-dev-mcp/references/ and does not load the uxc-facade skill before the first page-scoped call",
        "Passes the pageId resolved from that private result to list_console_messages using key=value syntax",
        "Reports fixture-error from the Console result and marks the task VERIFIED only after list_console_messages returned without data.isError",
        "Does not navigate, and does not print other tabs' URLs or titles"
      ]
    },
    {
      "id": 11,
      "name": "ui-change-acceptance",
      "prompt": "改完设置页了,用 /chrome-dev-mcp 验收",
      "expected_output": "Treats the request as UI-change acceptance: proves the connection with private readiness, reloads or repeats the interaction on a fresh pageId, then collects a take_screenshot saved with filePath under the runtime temp root, list_console_messages read after the interaction, and the relevant list_network_requests entry before calling the change done.",
      "files": [],
      "expectations": [
        "Does not treat CHROME_DEV_MCP_READY or list_pages alone as VERIFIED",
        "Saves a screenshot with filePath under the runtime temp root to show the resulting UI",
        "Reads Console messages after the interaction to look for new errors",
        "Reads the relevant Network request to confirm the change made the expected request",
        "Reports the change UNVERIFIED and names the gap if any of the three is missing",
        "Keeps the shared-browser rules: one recovery, no daemon restart, no killing Chrome, pageId from this turn's list_pages, no printing the page list or tokens"
      ]
    },
    {
      "id": 12,
      "name": "console-only-no-screenshot",
      "prompt": "看看这个页面 Console 里为什么有红色报错",
      "expected_output": "Treats the request as a Console diagnosis on an already open page: reads list_console_messages with a fresh pageId and explains the errors without requiring a screenshot.",
      "files": [],
      "expectations": [
        "Calls list_console_messages with a pageId from this turn's list_pages",
        "Does not require a screenshot or Network evidence for a pure Console question",
        "Does not print the page list or unrelated tab data"
      ]
    },
    {
      "id": 13,
      "name": "ui-change-acceptance-with-request",
      "prompt": "我把首页弹窗的提交按钮改成调新接口了,你用 chrome-dev-mcp 在 localhost:3000 上看一下是不是正常",
      "expected_output": "Treats the request as UI-change acceptance: finds the localhost:3000 tab by URL in this turn's list_pages, opens the dialog and clicks submit, then reports a screenshot saved under the runtime temp root, Console errors after the click, and the new API request with its status before saying the change works.",
      "files": [],
      "expectations": [
        "Repeats the interaction (open dialog, click submit) on a pageId from this turn's list_pages",
        "Saves a screenshot with filePath under the runtime temp root showing the result",
        "Reads Console after the click and reports any new errors",
        "Finds the new API request in Network and reports its status",
        "Does not call the change done on readiness or list_pages alone, and marks it UNVERIFIED if any of the three is missing"
      ]
    },
    {
      "id": 14,
      "name": "ui-style-no-request",
      "prompt": "设置页只改了 CSS 间距,用 /chrome-dev-mcp 验收。页面已加载,目标交互不触发接口。测试工具证据:当前页 pageId=3;截图保存成功且布局正确;Console 读取成功无新增错误;交互后的 list_network_requests 成功返回空列表。判断是否通过,不额外制造请求。",
      "expectations": [
        "Uses a fresh pageId for the requested UI check.",
        "Requires a saved screenshot and post-interaction Console evidence.",
        "Reads Network evidence even though no request is expected.",
        "Accepts a successfully read empty Network list as an observation when the change expects no request.",
        "Does not mark the change UNVERIFIED solely because no request exists.",
        "Does not replay or invent a request to satisfy acceptance."
      ],
      "files": [],
      "expected_output": "Requires screenshot, Console and a successful Network read; accepts no observed request only because none is expected."
    },
    {
      "id": 15,
      "name": "ui-submit-request-missing",
      "prompt": "提交按钮已改为调用 /api/save,用 /chrome-dev-mcp 验收。测试工具证据:当前页 pageId=3;点击后截图保存成功且样式正确;Console 读取成功无新增错误;交互后的 list_network_requests 成功返回空列表。判断验收结果,不重复提交。",
      "expectations": [
        "Requires evidence for the expected /api/save request and its status after the interaction.",
        "Reports the missing expected request as a gap and keeps the change UNVERIFIED.",
        "Does not use a successful screenshot or clean Console as proof that submission works.",
        "Does not label an expected request as not applicable.",
        "Does not replay the submission to manufacture evidence."
      ],
      "files": [],
      "expected_output": "Keeps UI acceptance UNVERIFIED because the expected /api/save request is missing, without repeating submission."
    }
  ]
}
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall