MCPInstall

browser-preview

Preview panel and iframe diagnostics: registry, ports, paths, blank screens. Use when a preview tab is broken or missing (e.g. white screen, tab disappeared, 404 on assets, preview not loading, list running services).

Installs
4.6K
Last 24h
—
Trending rank
—
Security
Pass

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/starchild-ai-agent/official-skills --skill browser-preview
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads browser-preview automatically when relevant.

  3. 03

    Review the result

    The agent follows the SKILL.md steps, scripts and references bundled with the skill.

Security audits

Independent scans of the skill's instructions and bundled files.

Gen Agent Trust Hub PassSAFE ·The skill provides instructions for diagnosing and fixing browser preview issues in a development environment. It includes logic for checking service registries, verifying local ports, and ensuring web assets use relative paths to work correctly behind a reverse proxy. No malicious patterns were detected.
Socket PassNo alerts
Snyk PassLOW ·Risk: LOW · No issues
Runlayer PassNONE ·1 file scanned · No issues
ZeroLeaks PassNONE ·Score: 93/100 · 2 sections analyzed

SKILL.md

The instructions your agent reads when the skill is triggered.

Browser Preview

You already know preview_serve and preview_stop. This skill fills the gap: what happens after preview_serve returns a URL — how the user actually sees it.

What is the Preview Panel

The frontend has a right-side panel with two tabs: Files and Preview. The Preview tab renders preview URLs inside an iframe. When you call preview_serve, the frontend automatically opens a Preview tab loading that URL. (Jobs are now managed in a standalone modal accessible from the left sidebar, not in the right panel.)

Key facts:

  • Each preview_serve call creates one Preview tab
  • URL format: https://<host>/preview/{id}/
  • Preview panel has a ⋮ menu (top-right) showing "RUNNING SERVICES" list
  • Preview tab can be closed by the user without stopping the backend service
  • Backend service stopping → Preview tab shows an error page

⚠️ CRITICAL: Never Tell Users to Access localhost

The user's browser CANNOT access localhost or 127.0.0.1. These addresses point to the server container, not the user's machine. The preview architecture uses a reverse proxy:

User's Browser → https://<host>/preview/{id}/path → (reverse proxy) → 127.0.0.1:{port}/path

Rules:

  • NEVER tell the user to visit http://localhost:{port} or http://127.0.0.1:{port} — they cannot reach it
  • ALWAYS direct users to the preview URL: /preview/{id}/ (or the full URL https://<host>/preview/{id}/)
  • curl http://localhost:{port} is for your own server-side diagnostics only — never suggest it to the user as a way to "test" the preview
  • When a preview is running, tell the user: "Check the Preview panel, or refresh the Preview panel"
  • If you need to give the user a URL, use the url field returned by preview_serve (format: /preview/{id}/)

⚠️ Static Assets Must Use Relative Paths

Because previews are served under /preview/{id}/, absolute paths in HTML/JS/CSS will break. The reverse proxy strips the /preview/{id} prefix before forwarding to the backend, but the browser resolves absolute paths from the domain root.

Example of the problem:

<!-- ❌ BROKEN: browser requests https://host/static/app.js → 404 (bypasses preview proxy) -->
<script src="/static/app.js"></script>

<!-- ✅ WORKS: browser requests https://host/preview/{id}/static/app.js → proxied correctly -->
<script src="static/app.js"></script>
<script src="./static/app.js"></script>

Common patterns to fix:

Broken (absolute)Fixed (relative)
"/static/app.js""static/app.js" or "./static/app.js"
"/api/users""api/users" or "./api/users"
"/images/logo.png""images/logo.png" or "./images/logo.png"
url('/fonts/x.woff')url('./fonts/x.woff')
fetch('/data.json')fetch('data.json')

Check ALL places where paths appear:

  1. HTML src, href attributes
  2. JavaScript fetch(), XMLHttpRequest, dynamic imports
  3. CSS url() references
  4. JavaScript string literals (e.g., '/static/' in template strings or concatenation)
  5. Framework config files (e.g., publicPath, base, assetPrefix)

⚠️ Be thorough — it's common to fix CSS url() but miss JS string literals like '/static/' (with single quotes). Search for ALL occurrences of absolute paths across all file types.

⚠️ Do NOT Browse Filesystem to Debug Previews

Never look at workspace directories like preview/, output/, or random folders to understand preview state. Those are user data, not preview service state.

The only sources of truth:

  1. Registry file: /data/previews.json (running services)
  2. History file: /data/preview_history.json (all past services)
  3. preview_serve / preview_stop tools
  4. Port checks via curl (server-side only, for your diagnostics)

Do NOT use ls/find on workspace directories to diagnose preview issues. Do NOT call unrelated tools like list_scheduled_tasks. Stay focused.

Step-by-Step: Diagnosing Preview Issues

When a user reports any Preview panel problem, follow this exact sequence:

Step 1: Read the registry (running services)

cat /data/previews.json 2>/dev/null || echo "NO_REGISTRY"

⚠️ Your bash CWD is /data/workspace/. The registry is at /data/previews.json (absolute path, one level up). Always use the absolute path.

JSON structure:

{
  "previews": [
    {"id": "f343befc", "title": "My App", "dir": "/data/workspace/my-project", "command": "npm run dev", "port": 9080, "is_builtin": false}
  ]
}

Step 2: Branch based on registry state

If registry has entries → Go to Step 3 (verify services) If registry is empty or missing → Go to Step 4 (check history)

Step 3: Registry has entries — verify and fix

For each preview in the registry, check if the port is responding server-side (this is your diagnostic, not for the user):

curl -s -o /dev/null -w "%{http_code}" http://localhost:{port}

If port responds (200):

  • The service IS running. Tell the user:
    • "You have a running service: {title}"
    • "Click the ⋮ menu at the top-right of the Preview panel, then click it in the RUNNING SERVICES list to reopen"
    • Preview URL: /preview/{id}/
  • If user says the ⋮ menu is empty or doesn't show the service → frontend lost sync. Fix by recreating: preview_stop(id) then preview_serve(dir, title, command) using the info from the registry. This forces the frontend to re-register the tab.

If port does NOT respond:

  • Process crashed but registry entry remains. Recreate:
    preview_stop(id="{id}")
    preview_serve(dir="{dir}", title="{title}", command="{command}")
    

Step 4: No running services — check history first, then scan workspace

When there are no running services, use a two-tier lookup to find projects the user can preview:

Tier 1: Read preview history (preferred — fast and accurate)

cat /data/preview_history.json 2>/dev/null || echo "NO_HISTORY"

JSON structure:

{
  "history": [
    {
      "id": "f343befc",
      "title": "Trading System",
      "dir": "/data/workspace/my-project",
      "command": "python main.py",
      "port": 8000,
      "is_builtin": false,
      "created_at": 1709100000.0,
      "last_started_at": 1709200000.0
    }
  ]
}

History entries are never removed by preview_stop — they persist across restarts. Entries are automatically pruned only when the project directory no longer exists.

If history has entries:

  • List all history entries to the user with title, directory, and last started time
  • Ask which one they want to restart
  • Call preview_serve with the dir, title, and command from the history entry

If user says a project is missing from history → fall through to Tier 2.

Tier 2: Scan workspace (fallback — when history is empty or incomplete)

find /data/workspace -maxdepth 2 \( -name "package.json" -o -name "index.html" -o -name "*.html" -o -name "app.py" -o -name "main.py" -o -name "vite.config.*" \) -not -path "*/node_modules/*" -not -path "*/skills/*" -not -path "*/memory/*" -not -path "*/prompt/*" -not -path "*/.git/*" 2>/dev/null

Then:

  1. List discovered projects with brief descriptions
  2. Ask the user which one to preview
  3. Call preview_serve with the appropriate directory

Don't just say "no services running" and stop. Always check history first, then scan, and offer options.

Quick Reference

User saysYou do
"tab disappeared" / "tab 不见了"Step 1 → 2 → 3 or 4
"blank page" / "白屏"Check port (server-side), if dead → recreate; if alive → check for absolute path issues
"not updating" / "内容没更新"Suggest refresh button in Preview tab, or recreate preview
"port conflict" / "端口冲突"preview_stop old → preview_serve new
"can't see service" / "⋮ menu empty"preview_stop + preview_serve to force re-register
"where's my project" / "what did I build"Read /data/preview_history.json and list entries
"resource load failed" / "JS/CSS 404"Check for absolute paths (/static/, /api/), fix to relative paths

What You Cannot Do

  • Cannot directly open/close/refresh Preview tabs (frontend UI)
  • Cannot force-refresh the iframe
  • Cannot read what the iframe displays

When you can't do something, tell the user the manual action (e.g., "click refresh in Preview tab"). If manual action doesn't work, recreate the preview with preview_stop + preview_serve.

Common Mistakes to Avoid

  1. ❌ Telling user to "visit http://localhost:18791/" — user cannot access localhost
  2. ❌ Saying "refresh the page at localhost" — meaningless to the user
  3. ❌ Only fixing CSS url() paths but missing JS string literals with absolute paths
  4. ❌ Forgetting to check ALL file types (HTML, JS, CSS, config) for absolute paths
  5. ✅ Always use /preview/{id}/ as the user-facing URL
  6. ✅ Always use curl localhost:{port} only for your own server-side diagnostics
  7. ✅ After fixing paths, call preview_stop + preview_serve to restart, then tell user to check Preview panel

FAQ

What does the browser-preview skill do?

Preview panel and iframe diagnostics: registry, ports, paths, blank screens. Use when a preview tab is broken or missing (e.g. white screen, tab disappeared, 404 on assets, preview not loading, list running services).

How do I install browser-preview?

Run npx skills add https://github.com/starchild-ai-agent/official-skills --skill browser-preview in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/browser-preview/ (personal) or .claude/skills/browser-preview/ (project).

Which AI agents can use browser-preview?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is browser-preview safe to install?

Security partners have reviewed it: Gen Agent Trust Hub passed; Socket passed; Snyk passed; Runlayer passed; ZeroLeaks passed. As with any skill, read the SKILL.md and bundled scripts before installing.

Who maintains browser-preview?

It is published by starchild-ai-agent in starchild-ai-agent/official-skills. It has 4,620 recorded installs.

More from starchild-ai-agent/official-skills

View all →

coinglass

starchild-ai-agent/official-skills

Crypto derivatives data: funding rates, open interest, liquidations, long/short ratios. Use when researching perp markets or comparing ETF flows (e.g. BTC funding, ETH OI). NOTE: platform API key is Startup tier — liquidation heatmap, coins-markets detail, liquidation order book and Hyperliquid positions require higher plans (see Plan Tiers).

Business & Finance11.5K installs

wallet

starchild-ai-agent/official-skills

Multi-chain wallet: EVM and Solana balances, transfers, signing, and policy. Use when checking balances, sending tokens, signing typed data, or proposing a wallet policy (e.g. send 10 USDC on Base, sign EIP-712, Solana balance).

AI & Agents8.9K installs

hyperliquid

starchild-ai-agent/official-skills

Trade perp futures, spot, and RWA on Hyperliquid DEX with up to asset max leverage. Use when placing perp or spot orders, setting TP/SL, or moving funds on Hyperliquid (e.g. long BTC 5x, sell ETH, deposit USDC, set stop).

AI & Agents8.7K installs

coingecko

starchild-ai-agent/official-skills

Crypto spot prices, OHLC charts, market discovery, and global stats. Use when looking up coin prices, market caps, trending coins, sector rankings, or comparing tokens (e.g. BTC price, ETH chart, top gainers).

AI & Agents7.9K installs

twitter

starchild-ai-agent/official-skills

Twitter/X data: fetch tweets, search, user profiles, followers, replies, trends. Use for any x.com or twitter.com URL or lookup (e.g. summarize this tweet, recent posts by @vitalikbuterin, search $SOL minfaves:50).

Marketing & SEO7.9K installs

skill-creator

starchild-ai-agent/official-skills

Scaffold new skills with valid frontmatter, directory layout, and a starter SKILL.md. Use when building a new reusable workflow or wrapping a new API (e.g. create a kalshi skill, scaffold an API helper, start a charting skill).

AI & Agents7.8K installs

Popular productivity & automation skills

View category →

agent-browser

vercel-labs/agent-browser

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.

OfficialProductivity & Automation1.3M installs

lark-doc

open.feishu.cn

飞书云文档(Docx / Wiki)内容操作:读取、创建、编辑文档,插入或下载图片附件,以及操作思维笔记。用户提供文档 URL/token(包括 doubao.com 的 /docx/、/wiki/)时使用;按 URL 路径/token 而非域名路由;/page/ 是妙搭应用,非飞书云文档。文档内嵌资源按读取参考中的统一规则分流。独立评论操作走 lark-drive;随正文读取评论使用 docs +fetch。表格或 Base 内部数据操作不在本 skill。

lark-base

open.feishu.cn

飞书多维表格(Base)操作:建表、字段、记录、视图、统计、公式/lookup、表单、仪表盘、应用模式(BaseApp/AppMode 页面与组件)、Workspace 目录、workflow、角色权限、模板中心(多维表格模板分类/列表/搜索);遇到 Base/多维表格/bitable、BaseApp/AppMode、/base/ 或 /app/ 链接时使用。BaseApp 不走 lark-apps;文件导入/导出转 lark-drive,认证/授权转 lark-shared。

lark-drive

open.feishu.cn

飞书云空间(云盘/云存储):管理 Drive 文件和文件夹,包含上传/下载、创建文件夹、复制/移动/删除、查看元数据、查询权限设置、评论/权限/订阅、标题、版本、飞书文档密级标签(secure labels)和本地文件导入。用户需要整理云盘目录、处理云空间资源 URL/token、判断链接类型/真实 token/标题,或导入 Word/Markdown/Excel/CSV/PPTX/.base 为 docx/sheet/bitable/slides 时使用;doubao.com 云空间 URL/token 也按资源路径和 token 路由,不回退 WebFetch。不负责:文档内容编辑(走 lark-doc)、表格/Base 表内数据操作(走 lark-sheets/lark-base)、知识空间节点/成员管理(走 lark-wiki)、原生 Markdown 文件读写/patch/diff(走 lark-markdown)。

lark-shared

open.feishu.cn

Use for lark-cli setup/auth tasks: auth login/status/logout, user vs bot identity, business-domain permissions (--domain, including all/docs/drive), missing scopes, revoking authorization, or handling notice JSON.

lark-im

open.feishu.cn

飞书即时通讯:收发消息和管理群聊。发送和回复消息、搜索聊天记录、管理群聊成员、上传下载图片和文件、管理表情回复、发送应用内/短信/电话加急、发送和处理交互卡片(Interactive Card)、监听卡片按钮回调(card.action.trigger)。当用户需要发消息、查看或搜索聊天记录、下载聊天中的文件、查看群成员、搜索群、创建群聊或话题群、管理标记数据、管理 Feed 置顶(添加/移除/查询置顶会话)、管理标签数据、处理卡片回调时使用。