8
4

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

2026年!初アウトプットをしよう
@tarou-imokenpi(たろう)

Vercel製 agent-browserの使い方

8
Posted at

agent-browserとは

agent-browserは、AIエージェントがWebを操作するために設計されたCLIブラウザツールです。

このツールの特徴は、Webページの構造をアクセシビリティツリーとして取得し、各要素に参照ID（@e1など）を付与することで、エージェントがその IDを使ってクリック・入力・取得を直感的に行えることです。

CLIはRust言語で開発されているため、Node.js製のツールと比べて起動速度が速く、Agent Skillsにも対応しているため簡単に導入できます。

Agent Skillsにも対応しており簡単に導入可能でこれについても解説します。

インストール

公式リポジトリ

npmでのインストールが推奨されています。初回インストール後、agent-browser installでChromiumが自動的にダウンロードされます。

npm install -g agent-browser
agent-browser install

動作確認&コマンドの確認

agent-browser -h

agent-browser - fast browser automation CLI for AI agents

Usage: agent-browser <command> [args] [options]

Core Commands:
  open <url>                 Navigate to URL
  click <sel>                Click element (or @ref)
  dblclick <sel>             Double-click element
  type <sel> <text>          Type into element
  fill <sel> <text>          Clear and fill
  press <key>                Press key (Enter, Tab, Control+a)
  hover <sel>                Hover element
  focus <sel>                Focus element
  check <sel>                Check checkbox
  uncheck <sel>              Uncheck checkbox
  select <sel> <val>         Select dropdown option
  drag <src> <dst>           Drag and drop
  upload <sel> <files...>    Upload files
  scroll <dir> [px]          Scroll (up/down/left/right)
  scrollintoview <sel>       Scroll element into view
  wait <sel|ms>              Wait for element or time
  screenshot [path]          Take screenshot
  pdf <path>                 Save as PDF
  snapshot                   Accessibility tree with refs (for AI)
  eval <js>                  Run JavaScript
  close                      Close browser

Navigation:
  back                       Go back
  forward                    Go forward
  reload                     Reload page

Get Info:  agent-browser get <what> [selector]
  text, html, value, attr <name>, title, url, count, box

Check State:  agent-browser is <what> <selector>
  visible, enabled, checked

Find Elements:  agent-browser find <locator> <value> <action> [text]
  role, text, label, placeholder, alt, title, testid, first, last, nth

Mouse:  agent-browser mouse <action> [args]
  move <x> <y>, down [btn], up [btn], wheel <dy> [dx]

Browser Settings:  agent-browser set <setting> [value]
  viewport <w> <h>, device <name>, geo <lat> <lng>
  offline [on|off], headers <json>, credentials <user> <pass>
  media [dark|light] [reduced-motion]

Network:  agent-browser network <action>
  route <url> [--abort|--body <json>]
  unroute [url]
  requests [--clear] [--filter <pattern>]

Storage:
  cookies [get|set|clear]    Manage cookies
  storage <local|session>    Manage web storage

Tabs:
  tab [new|list|close|<n>]   Manage tabs

Debug:
  trace start|stop [path]    Record trace
  console [--clear]          View console logs
  errors [--clear]           View page errors
  highlight <sel>            Highlight element

Sessions:
  session                    Show current session name
  session list               List active sessions

Setup:
  install                    Install browser binaries
  install --with-deps        Also install system dependencies (Linux)

Snapshot Options:
  -i, --interactive          Only interactive elements
  -c, --compact              Remove empty structural elements
  -d, --depth <n>            Limit tree depth
  -s, --selector <sel>       Scope to CSS selector

Options:
  --session <name>           Isolated session (or AGENT_BROWSER_SESSION env)
  --json                     JSON output
  --full, -f                 Full page screenshot
  --headed                   Show browser window (not headless)
  --debug                    Debug output

Examples:
  agent-browser open example.com
  agent-browser snapshot -i              # Interactive elements only
  agent-browser click @e2                # Click by ref from snapshot
  agent-browser fill @e3 "test@example.com"
  agent-browser find role button click --name Submit
  agent-browser get text @e1
  agent-browser screenshot --full

基本的な使い方

agent-browser open example.com
agent-browser snapshot
agent-browser click @e2
agent-browser fill @e3 "test@example.com"
agent-browser get text @e1
agent-browser screenshot page.png
agent-browser close

まずはopen [url]でブラウザ起動&URLにアクセスします。
次にsnapshotを実行します、これで要素やリンクに対して以下のような様々なコマンドが使用できます。

そして、終了時はcloseでブラウザを終了しておきましょう。

snapshot とは

snapshotコマンドはページのアクセシビリティツリーを取得し、要素への参照（refs）を付与してAIエージェントが要素を確実に操作できるようにする機能です。
PlaywrightのariaSnapshot()でアクセシビリティツリーを取得しているようです。

CLI コマンド一覧(抜粋)

カテゴリ コマンド 引数 説明
ナビゲーション open <url> URLに移動する
back - 履歴を戻る
forward - 履歴を進む
reload - ページを再読み込み
基本操作 click <selector> 要素をクリック
dblclick <selector> 要素をダブルクリック
fill <selector> <text> 要素をクリアして入力
type <selector> <text> 要素にテキストを入力
hover <selector> 要素にホバー
focus <selector> 要素にフォーカス
check <selector> チェックボックスをオン
uncheck <selector> チェックボックスをオフ
select <selector> <value> ドロップダウンを選択
drag <source> <target> ドラッグ＆ドロップ
upload <selector> <files...> ファイルをアップロード
キーボード press <key> キーを押す
keydown <key> キーを押し続ける
keyup <key> キーを離す
スクロール scroll <direction> [amount] ページをスクロール
scrollintoview <selector> 要素をビューにスクロール
待機 wait <selector|ms> 要素または時間を待つ
スクリーンショット/PDF screenshot [path] スクリーンショットを撮る
pdf <path> PDFとして保存
スナップショット snapshot [options] アクセシビリティツリーを取得
評価 eval <javascript> JavaScriptを実行
情報取得 get <what> [selector] 情報を取得（text, html, value, attr, url, title, count, box）
終了 close - ブラウザを閉じる

スナップショットオプション

オプション 説明
-i, --interactive 対話的な要素のみ表示（ボタン、リンク、入力欄）
-c, --compact 空の構造要素を削除
-d, --depth <n> ツリーの深さを制限
-s, --selector <sel> CSSセレクターで範囲を指定

グローバルオプション

オプション 説明
--session <name> 分離されたセッションを使用
--json JSON出力
--full, -f ページ全体のスクリーンショット
--headed ブラウザウィンドウを表示
--debug デバッグ出力
  • 詳細なヘルプは agent-browser --help で表示されます
  • セマンティックロケーター（find コマンド）には role, text, label, placeholder, alt, title, testid, first, last, nth などが使用できます

実際にsnapshotを実行した例

agent-browser open https://www.anthropic.com/engineering
agent-browser snapshot
出力 
- document:
  - banner:
    - link "Skip to main content" [ref=e1]:
      - /url: "#main-content"
    - link "Skip to footer" [ref=e2]:
      - /url: "#footer"
    - link "Home" [ref=e3]:
      - /url: /
      - img
    - navigation:
      - list:
        - listitem:
          - link "Research" [ref=e4]:
            - /url: /research
        - listitem:
          - link "Economic Futures" [ref=e5]:
            - /url: /economic-futures
        - listitem:
          - button "Commitments" [ref=e6]:
            - text: Commitments
            - img
        - listitem:
          - button "Learn" [ref=e7]:
            - text: Learn
            - img
        - listitem:
          - link "News" [ref=e8]:
            - /url: /news
    - link "Try Claude" [ref=e9]:
      - /url: https://claude.ai/
    - img
  - main:
    - article:
      - 'heading "Engineering at Anthropic: Inside the team building reliable AI systems" [level=2]'
      - link "Start building" [ref=e10]:
        - /url: https://platform.claude.com/
      - link "Developer docs" [ref=e11]:
        - /url: https://www.anthropic.com/docs
      - img "Demystifying evals for AI agents"
      - article:
        - link "Featured Demystifying evals for AI agents The capabilities that make agents useful also make them difficult to evaluate. The strategies that work across deployments combine techniques to match the complexity of the systems they measure." [ref=e12]:
          - /url: /engineering/demystifying-evals-for-ai-agents
          - text: Featured
          - heading "Demystifying evals for AI agents" [ref=e13] [level=2]
          - paragraph: The capabilities that make agents useful also make them difficult to evaluate. The strategies that work across deployments combine techniques to match the complexity of the systems they measure.
      - article:
        - link "Effective harnesses for long-running agents Nov 26, 2025" [ref=e14]:
          - /url: /engineering/effective-harnesses-for-long-running-agents
          - heading "Effective harnesses for long-running agents" [ref=e15] [level=3]
          - text: Nov 26, 2025
      - article:
        - link "Introducing advanced tool use on the Claude Developer Platform Nov 24, 2025" [ref=e16]:
          - /url: /engineering/advanced-tool-use
          - heading "Introducing advanced tool use on the Claude Developer Platform" [ref=e17] [level=3]
          - text: Nov 24, 2025
      - article:
        - 'link "Code execution with MCP: Building more efficient agents Nov 04, 2025"':
          - /url: /engineering/code-execution-with-mcp
          - 'heading "Code execution with MCP: Building more efficient agents" [level=3]'
          - text: Nov 04, 2025
      - article:
        - 'link "Beyond permission prompts: making Claude Code more secure and autonomous Oct 20, 2025"':
          - /url: /engineering/claude-code-sandboxing
          - 'heading "Beyond permission prompts: making Claude Code more secure and autonomous" [level=3]'
          - text: Oct 20, 2025
      - article:
        - link "Equipping agents for the real world with Agent Skills Oct 16, 2025" [ref=e18]:
          - /url: /engineering/equipping-agents-for-the-real-world-with-agent-skills
          - heading "Equipping agents for the real world with Agent Skills" [ref=e19] [level=3]
          - text: Oct 16, 2025
      - article:
        - link "Building agents with the Claude Agent SDK Sep 29, 2025" [ref=e20]:
          - /url: /engineering/building-agents-with-the-claude-agent-sdk
          - heading "Building agents with the Claude Agent SDK" [ref=e21] [level=3]
          - text: Sep 29, 2025
      - article:
        - link "Effective context engineering for AI agents Sep 29, 2025" [ref=e22]:
          - /url: /engineering/effective-context-engineering-for-ai-agents
          - heading "Effective context engineering for AI agents" [ref=e23] [level=3]
          - text: Sep 29, 2025
      - article:
        - link "A postmortem of three recent issues Sep 17, 2025" [ref=e24]:
          - /url: /engineering/a-postmortem-of-three-recent-issues
          - heading "A postmortem of three recent issues" [ref=e25] [level=3]
          - text: Sep 17, 2025
      - article:
        - link "Writing effective tools for agents — with agents Sep 11, 2025" [ref=e26]:
          - /url: /engineering/writing-tools-for-agents
          - heading "Writing effective tools for agents — with agents" [ref=e27] [level=3]
          - text: Sep 11, 2025
      - article:
        - 'link "Desktop Extensions: One-click MCP server installation for Claude Desktop Jun 26, 2025"':
          - /url: /engineering/desktop-extensions
          - 'heading "Desktop Extensions: One-click MCP server installation for Claude Desktop" [level=3]'
          - text: Jun 26, 2025
      - article:
        - link "How we built our multi-agent research system Jun 13, 2025" [ref=e28]:
          - /url: /engineering/multi-agent-research-system
          - heading "How we built our multi-agent research system" [ref=e29] [level=3]
          - text: Jun 13, 2025
      - article:
        - 'link "Claude Code: Best practices for agentic coding Apr 18, 2025"':
          - /url: /engineering/claude-code-best-practices
          - 'heading "Claude Code: Best practices for agentic coding" [level=3]'
          - text: Apr 18, 2025
      - article:
        - 'link "The \"think\" tool: Enabling Claude to stop and think in complex tool use situations Mar 20, 2025"':
          - /url: /engineering/claude-think-tool
          - 'heading "The \"think\" tool: Enabling Claude to stop and think in complex tool use situations" [level=3]'
          - text: Mar 20, 2025
      - article:
        - link "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet Jan 06, 2025" [ref=e30]:
          - /url: /engineering/swe-bench-sonnet
          - heading "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet" [ref=e31] [level=3]
          - text: Jan 06, 2025
      - article:
        - link "Building effective agents Dec 19, 2024" [ref=e32]:
          - /url: /engineering/building-effective-agents
          - heading "Building effective agents" [ref=e33] [level=3]
          - text: Dec 19, 2024
      - article:
        - link "Introducing Contextual Retrieval Sep 19, 2024" [ref=e34]:
          - /url: /engineering/contextual-retrieval
          - heading "Introducing Contextual Retrieval" [ref=e35] [level=3]
          - text: Sep 19, 2024
      - paragraph: Want to help us build the future of safe AI?
      - link "See open roles" [ref=e36]:
        - /url: /jobs
        - text: See open roles
        - img
  - contentinfo "Site footer":
    - link "Return to homepage" [ref=e37]:
      - /url: /
      - img
    - navigation "Footer navigation" [ref=e38]:
      - heading "Products" [ref=e39] [level=3]
      - list:
        - listitem:
          - link "Claude" [ref=e40]:
            - /url: https://claude.com/product/overview
        - listitem:
          - link "Claude Code" [ref=e41]:
            - /url: https://claude.com/product/claude-code
        - listitem:
          - link "Claude in Chrome" [ref=e42]:
            - /url: https://claude.com/chrome
        - listitem:
          - link "Claude in Excel" [ref=e43]:
            - /url: https://claude.com/claude-in-excel
        - listitem:
          - link "Claude in Slack" [ref=e44]:
            - /url: https://claude.com/claude-in-slack
        - listitem:
          - link "Skills" [ref=e45]:
            - /url: https://www.claude.com/skills
        - listitem:
          - link "Max plan" [ref=e46]:
            - /url: https://claude.com/pricing/max
        - listitem:
          - link "Team plan" [ref=e47]:
            - /url: https://claude.com/pricing/team
        - listitem:
          - link "Enterprise plan" [ref=e48]:
            - /url: https://claude.com/pricing/enterprise
        - listitem:
          - link "Download app" [ref=e49]:
            - /url: https://claude.ai/download
        - listitem:
          - link "Pricing" [ref=e50]:
            - /url: https://claude.com/pricing
        - listitem:
          - link "Log in to Claude" [ref=e51]:
            - /url: https://claude.ai/
      - heading "Models" [ref=e52] [level=3]
      - list:
        - listitem:
          - link "Opus" [ref=e53]:
            - /url: https://www.anthropic.com/claude/opus
        - listitem:
          - link "Sonnet" [ref=e54]:
            - /url: https://www.anthropic.com/claude/sonnet
        - listitem:
          - link "Haiku" [ref=e55]:
            - /url: https://www.anthropic.com/claude/haiku
      - heading "Solutions" [ref=e56] [level=3]
      - list:
        - listitem:
          - link "AI agents" [ref=e57]:
            - /url: https://claude.com/solutions/agents
        - listitem:
          - link "Code modernization" [ref=e58]:
            - /url: https://claude.com/solutions/code-modernization
        - listitem:
          - link "Coding" [ref=e59]:
            - /url: https://claude.com/solutions/coding
        - listitem:
          - link "Customer support" [ref=e60]:
            - /url: https://claude.com/solutions/customer-support
        - listitem:
          - link "Education" [ref=e61]:
            - /url: https://claude.com/solutions/education
        - listitem:
          - link "Financial services" [ref=e62]:
            - /url: https://claude.com/solutions/financial-services
        - listitem:
          - link "Government" [ref=e63]:
            - /url: https://claude.com/solutions/government
        - listitem:
          - link "Healthcare" [ref=e64]:
            - /url: https://claude.com/solutions/healthcare
        - listitem:
          - link "Life sciences" [ref=e65]:
            - /url: https://claude.com/solutions/life-sciences
        - listitem:
          - link "Nonprofits" [ref=e66]:
            - /url: https://claude.com/solutions/nonprofits
      - heading "Claude Developer Platform" [ref=e67] [level=3]
      - list:
        - listitem:
          - link "Overview" [ref=e68]:
            - /url: https://claude.com/platform/api
        - listitem:
          - link "Developer docs" [ref=e69] [nth=1]:
            - /url: https://platform.claude.com/docs
        - listitem:
          - link "Pricing" [ref=e70] [nth=1]:
            - /url: https://claude.com/pricing#api
        - listitem:
          - link "Regional Compliance" [ref=e71]:
            - /url: https://claude.com/regional-compliance
        - listitem:
          - link "Amazon Bedrock" [ref=e72]:
            - /url: https://claude.com/partners/amazon-bedrock
        - listitem:
          - link "Google Cloud’s Vertex AI" [ref=e73]:
            - /url: https://claude.com/partners/google-cloud-vertex-ai
        - listitem:
          - link "Console login" [ref=e74]:
            - /url: https://platform.claude.com/
      - heading "Learn" [ref=e75] [level=3]
      - list:
        - listitem:
          - link "Blog" [ref=e76]:
            - /url: https://claude.com/blog
        - listitem:
          - link "Claude partner network" [ref=e77]:
            - /url: https://claude.com/partners
        - listitem:
          - link "Connectors" [ref=e78]:
            - /url: https://claude.com/connectors
        - listitem:
          - link "Courses" [ref=e79]:
            - /url: /learn
        - listitem:
          - link "Customer stories" [ref=e80]:
            - /url: https://claude.com/customers
        - listitem:
          - link "Engineering at Anthropic" [ref=e81]:
            - /url: /engineering
        - listitem:
          - link "Events" [ref=e82]:
            - /url: /events
        - listitem:
          - link "Powered by Claude" [ref=e83]:
            - /url: https://claude.com/partners/powered-by-claude
        - listitem:
          - link "Service partners" [ref=e84]:
            - /url: https://claude.com/partners/services
        - listitem:
          - link "Startups program" [ref=e85]:
            - /url: https://claude.com/programs/startups
        - listitem:
          - link "Tutorials" [ref=e86]:
            - /url: https://claude.com/resources/tutorials
        - listitem:
          - link "Use cases" [ref=e87]:
            - /url: https://claude.com/resources/use-cases
      - heading "Company" [ref=e88] [level=3]
      - list:
        - listitem:
          - link "Anthropic" [ref=e89]:
            - /url: /company
        - listitem:
          - link "Careers" [ref=e90]:
            - /url: /careers
        - listitem:
          - link "Economic Futures" [ref=e91] [nth=1]:
            - /url: /economic-index
        - listitem:
          - link "Research" [ref=e92] [nth=1]:
            - /url: /research
        - listitem:
          - link "News" [ref=e93] [nth=1]:
            - /url: /news
        - listitem:
          - link "Responsible Scaling Policy" [ref=e94]:
            - /url: https://www.anthropic.com/news/announcing-our-updated-responsible-scaling-policy
        - listitem:
          - link "Security and compliance" [ref=e95]:
            - /url: https://trust.anthropic.com/
        - listitem:
          - link "Transparency" [ref=e96]:
            - /url: /transparency
      - heading "Help and security" [ref=e97] [level=3]
      - list:
        - listitem:
          - link "Availability" [ref=e98]:
            - /url: https://www.anthropic.com/supported-countries
        - listitem:
          - link "Status" [ref=e99]:
            - /url: https://status.anthropic.com/
        - listitem:
          - link "Support center" [ref=e100]:
            - /url: https://support.claude.com/en/
      - heading "Terms and policies" [ref=e101] [level=3]
      - list:
        - button "Privacy choices" [ref=e102]
        - listitem:
          - link "Privacy policy" [ref=e103]:
            - /url: https://www.anthropic.com/legal/privacy
        - listitem:
          - link "Consumer health data privacy policy" [ref=e104]:
            - /url: https://www.anthropic.com/legal/consumer-health-data-privacy-policy
        - listitem:
          - link "Responsible disclosure policy" [ref=e105]:
            - /url: https://www.anthropic.com/responsible-disclosure-policy
        - listitem:
          - 'link "Terms of service: Commercial"':
            - /url: https://www.anthropic.com/legal/commercial-terms
        - listitem:
          - 'link "Terms of service: Consumer"':
            - /url: https://www.anthropic.com/legal/consumer-terms
        - listitem:
          - link "Usage policy" [ref=e106]:
            - /url: https://www.anthropic.com/legal/aup
    - contentinfo: © 2025 Anthropic PBC
    - navigation "Social media links" [ref=e107]:
      - listitem:
        - link "Visit our LinkedIn page" [ref=e108]:
          - /url: https://www.linkedin.com/company/anthropicresearch
          - img
      - listitem:
        - link "Visit our X (formerly Twitter) profile" [ref=e109]:
          - /url: https://x.com/AnthropicAI
          - img
      - listitem:
        - link "Visit our YouTube channel" [ref=e110]:
          - /url: https://www.youtube.com/@anthropic-ai
          - img
  - alert

agent-browser snapshot -i
出力 
- link "Skip to main content" [ref=e1]
- link "Skip to footer" [ref=e2]
- link "Home" [ref=e3]
- link "Research" [ref=e4]
- link "Economic Futures" [ref=e5]
- button "Commitments" [ref=e6]
- button "Learn" [ref=e7]
- link "News" [ref=e8]
- link "Try Claude" [ref=e9]
- link "Start building" [ref=e10]
- link "Developer docs" [ref=e11]
- link "Featured Demystifying evals for AI agents The capabilities that make agents useful also make them difficult to evaluate. The strategies that work across deployments combine techniques to match the complexity of the systems they measure." [ref=e12]
- link "Effective harnesses for long-running agents Nov 26, 2025" [ref=e13]
- link "Introducing advanced tool use on the Claude Developer Platform Nov 24, 2025" [ref=e14]
- link "Equipping agents for the real world with Agent Skills Oct 16, 2025" [ref=e15]
- link "Building agents with the Claude Agent SDK Sep 29, 2025" [ref=e16]
- link "Effective context engineering for AI agents Sep 29, 2025" [ref=e17]
- link "A postmortem of three recent issues Sep 17, 2025" [ref=e18]
- link "Writing effective tools for agents — with agents Sep 11, 2025" [ref=e19]
- link "How we built our multi-agent research system Jun 13, 2025" [ref=e20]
- link "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet Jan 06, 2025" [ref=e21]
- link "Building effective agents Dec 19, 2024" [ref=e22]
- link "Introducing Contextual Retrieval Sep 19, 2024" [ref=e23]
- link "See open roles" [ref=e24]
- link "Return to homepage" [ref=e25]
- link "Claude" [ref=e26]
- link "Claude Code" [ref=e27]
- link "Claude in Chrome" [ref=e28]
- link "Claude in Excel" [ref=e29]
- link "Claude in Slack" [ref=e30]
- link "Skills" [ref=e31]
- link "Max plan" [ref=e32]
- link "Team plan" [ref=e33]
- link "Enterprise plan" [ref=e34]
- link "Download app" [ref=e35]
- link "Pricing" [ref=e36]
- link "Log in to Claude" [ref=e37]
- link "Opus" [ref=e38]
- link "Sonnet" [ref=e39]
- link "Haiku" [ref=e40]
- link "AI agents" [ref=e41]
- link "Code modernization" [ref=e42]
- link "Coding" [ref=e43]
- link "Customer support" [ref=e44]
- link "Education" [ref=e45]
- link "Financial services" [ref=e46]
- link "Government" [ref=e47]
- link "Healthcare" [ref=e48]
- link "Life sciences" [ref=e49]
- link "Nonprofits" [ref=e50]
- link "Overview" [ref=e51]
- link "Developer docs" [ref=e52] [nth=1]
- link "Pricing" [ref=e53] [nth=1]
- link "Regional Compliance" [ref=e54]
- link "Amazon Bedrock" [ref=e55]
- link "Google Cloud’s Vertex AI" [ref=e56]
- link "Console login" [ref=e57]
- link "Blog" [ref=e58]
- link "Claude partner network" [ref=e59]
- link "Connectors" [ref=e60]
- link "Courses" [ref=e61]
- link "Customer stories" [ref=e62]
- link "Engineering at Anthropic" [ref=e63]
- link "Events" [ref=e64]
- link "Powered by Claude" [ref=e65]
- link "Service partners" [ref=e66]
- link "Startups program" [ref=e67]
- link "Tutorials" [ref=e68]
- link "Use cases" [ref=e69]
- link "Anthropic" [ref=e70]
- link "Careers" [ref=e71]
- link "Economic Futures" [ref=e72] [nth=1]
- link "Research" [ref=e73] [nth=1]
- link "News" [ref=e74] [nth=1]
- link "Responsible Scaling Policy" [ref=e75]
- link "Security and compliance" [ref=e76]
- link "Transparency" [ref=e77]
- link "Availability" [ref=e78]
- link "Status" [ref=e79]
- link "Support center" [ref=e80]
- button "Privacy choices" [ref=e81]
- link "Privacy policy" [ref=e82]
- link "Consumer health data privacy policy" [ref=e83]
- link "Responsible disclosure policy" [ref=e84]
- link "Usage policy" [ref=e85]
- link "Visit our LinkedIn page" [ref=e86]
- link "Visit our X (formerly Twitter) profile" [ref=e87]
- link "Visit our YouTube channel" [ref=e88]

agent-browser snapshot -c
出力 
  - banner:
    - link "Skip to main content" [ref=e1]:
      - /url: "#main-content"
    - link "Skip to footer" [ref=e2]:
      - /url: "#footer"
    - link "Home" [ref=e3]:
      - /url: /
    - navigation:
        - listitem:
          - link "Research" [ref=e4]:
            - /url: /research
        - listitem:
          - link "Economic Futures" [ref=e5]:
            - /url: /economic-futures
        - listitem:
          - button "Commitments" [ref=e6]:
            - text: Commitments
        - listitem:
          - button "Learn" [ref=e7]:
            - text: Learn
        - listitem:
          - link "News" [ref=e8]:
            - /url: /news
    - link "Try Claude" [ref=e9]:
      - /url: https://claude.ai/
  - main:
    - article:
      - 'heading "Engineering at Anthropic: Inside the team building reliable AI systems" [level=2]'
      - link "Start building" [ref=e10]:
        - /url: https://platform.claude.com/
      - link "Developer docs" [ref=e11]:
        - /url: https://www.anthropic.com/docs
      - article:
        - link "Featured Demystifying evals for AI agents The capabilities that make agents useful also make them difficult to evaluate. The strategies that work across deployments combine techniques to match the complexity of the systems they measure." [ref=e12]:
          - /url: /engineering/demystifying-evals-for-ai-agents
          - text: Featured
          - heading "Demystifying evals for AI agents" [ref=e13] [level=2]
          - paragraph: The capabilities that make agents useful also make them difficult to evaluate. The strategies that work across deployments combine techniques to match the complexity of the systems they measure.
      - article:
        - link "Effective harnesses for long-running agents Nov 26, 2025" [ref=e14]:
          - /url: /engineering/effective-harnesses-for-long-running-agents
          - heading "Effective harnesses for long-running agents" [ref=e15] [level=3]
          - text: Nov 26, 2025
      - article:
        - link "Introducing advanced tool use on the Claude Developer Platform Nov 24, 2025" [ref=e16]:
          - /url: /engineering/advanced-tool-use
          - heading "Introducing advanced tool use on the Claude Developer Platform" [ref=e17] [level=3]
          - text: Nov 24, 2025
          - /url: /engineering/code-execution-with-mcp
          - 'heading "Code execution with MCP: Building more efficient agents" [level=3]'
          - text: Nov 04, 2025
          - /url: /engineering/claude-code-sandboxing
          - 'heading "Beyond permission prompts: making Claude Code more secure and autonomous" [level=3]'
          - text: Oct 20, 2025
      - article:
        - link "Equipping agents for the real world with Agent Skills Oct 16, 2025" [ref=e18]:
          - /url: /engineering/equipping-agents-for-the-real-world-with-agent-skills
          - heading "Equipping agents for the real world with Agent Skills" [ref=e19] [level=3]
          - text: Oct 16, 2025
      - article:
        - link "Building agents with the Claude Agent SDK Sep 29, 2025" [ref=e20]:
          - /url: /engineering/building-agents-with-the-claude-agent-sdk
          - heading "Building agents with the Claude Agent SDK" [ref=e21] [level=3]
          - text: Sep 29, 2025
      - article:
        - link "Effective context engineering for AI agents Sep 29, 2025" [ref=e22]:
          - /url: /engineering/effective-context-engineering-for-ai-agents
          - heading "Effective context engineering for AI agents" [ref=e23] [level=3]
          - text: Sep 29, 2025
      - article:
        - link "A postmortem of three recent issues Sep 17, 2025" [ref=e24]:
          - /url: /engineering/a-postmortem-of-three-recent-issues
          - heading "A postmortem of three recent issues" [ref=e25] [level=3]
          - text: Sep 17, 2025
      - article:
        - link "Writing effective tools for agents — with agents Sep 11, 2025" [ref=e26]:
          - /url: /engineering/writing-tools-for-agents
          - heading "Writing effective tools for agents — with agents" [ref=e27] [level=3]
          - text: Sep 11, 2025
          - /url: /engineering/desktop-extensions
          - 'heading "Desktop Extensions: One-click MCP server installation for Claude Desktop" [level=3]'
          - text: Jun 26, 2025
      - article:
        - link "How we built our multi-agent research system Jun 13, 2025" [ref=e28]:
          - /url: /engineering/multi-agent-research-system
          - heading "How we built our multi-agent research system" [ref=e29] [level=3]
          - text: Jun 13, 2025
          - /url: /engineering/claude-code-best-practices
          - 'heading "Claude Code: Best practices for agentic coding" [level=3]'
          - text: Apr 18, 2025
          - /url: /engineering/claude-think-tool
          - 'heading "The \"think\" tool: Enabling Claude to stop and think in complex tool use situations" [level=3]'
          - text: Mar 20, 2025
      - article:
        - link "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet Jan 06, 2025" [ref=e30]:
          - /url: /engineering/swe-bench-sonnet
          - heading "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet" [ref=e31] [level=3]
          - text: Jan 06, 2025
      - article:
        - link "Building effective agents Dec 19, 2024" [ref=e32]:
          - /url: /engineering/building-effective-agents
          - heading "Building effective agents" [ref=e33] [level=3]
          - text: Dec 19, 2024
      - article:
        - link "Introducing Contextual Retrieval Sep 19, 2024" [ref=e34]:
          - /url: /engineering/contextual-retrieval
          - heading "Introducing Contextual Retrieval" [ref=e35] [level=3]
          - text: Sep 19, 2024
      - paragraph: Want to help us build the future of safe AI?
      - link "See open roles" [ref=e36]:
        - /url: /jobs
        - text: See open roles
  - contentinfo "Site footer":
    - link "Return to homepage" [ref=e37]:
      - /url: /
    - navigation "Footer navigation" [ref=e38]:
      - heading "Products" [ref=e39] [level=3]
        - listitem:
          - link "Claude" [ref=e40]:
            - /url: https://claude.com/product/overview
        - listitem:
          - link "Claude Code" [ref=e41]:
            - /url: https://claude.com/product/claude-code
        - listitem:
          - link "Claude in Chrome" [ref=e42]:
            - /url: https://claude.com/chrome
        - listitem:
          - link "Claude in Excel" [ref=e43]:
            - /url: https://claude.com/claude-in-excel
        - listitem:
          - link "Claude in Slack" [ref=e44]:
            - /url: https://claude.com/claude-in-slack
        - listitem:
          - link "Skills" [ref=e45]:
            - /url: https://www.claude.com/skills
        - listitem:
          - link "Max plan" [ref=e46]:
            - /url: https://claude.com/pricing/max
        - listitem:
          - link "Team plan" [ref=e47]:
            - /url: https://claude.com/pricing/team
        - listitem:
          - link "Enterprise plan" [ref=e48]:
            - /url: https://claude.com/pricing/enterprise
        - listitem:
          - link "Download app" [ref=e49]:
            - /url: https://claude.ai/download
        - listitem:
          - link "Pricing" [ref=e50]:
            - /url: https://claude.com/pricing
        - listitem:
          - link "Log in to Claude" [ref=e51]:
            - /url: https://claude.ai/
      - heading "Models" [ref=e52] [level=3]
        - listitem:
          - link "Opus" [ref=e53]:
            - /url: https://www.anthropic.com/claude/opus
        - listitem:
          - link "Sonnet" [ref=e54]:
            - /url: https://www.anthropic.com/claude/sonnet
        - listitem:
          - link "Haiku" [ref=e55]:
            - /url: https://www.anthropic.com/claude/haiku
      - heading "Solutions" [ref=e56] [level=3]
        - listitem:
          - link "AI agents" [ref=e57]:
            - /url: https://claude.com/solutions/agents
        - listitem:
          - link "Code modernization" [ref=e58]:
            - /url: https://claude.com/solutions/code-modernization
        - listitem:
          - link "Coding" [ref=e59]:
            - /url: https://claude.com/solutions/coding
        - listitem:
          - link "Customer support" [ref=e60]:
            - /url: https://claude.com/solutions/customer-support
        - listitem:
          - link "Education" [ref=e61]:
            - /url: https://claude.com/solutions/education
        - listitem:
          - link "Financial services" [ref=e62]:
            - /url: https://claude.com/solutions/financial-services
        - listitem:
          - link "Government" [ref=e63]:
            - /url: https://claude.com/solutions/government
        - listitem:
          - link "Healthcare" [ref=e64]:
            - /url: https://claude.com/solutions/healthcare
        - listitem:
          - link "Life sciences" [ref=e65]:
            - /url: https://claude.com/solutions/life-sciences
        - listitem:
          - link "Nonprofits" [ref=e66]:
            - /url: https://claude.com/solutions/nonprofits
      - heading "Claude Developer Platform" [ref=e67] [level=3]
        - listitem:
          - link "Overview" [ref=e68]:
            - /url: https://claude.com/platform/api
        - listitem:
          - link "Developer docs" [ref=e69] [nth=1]:
            - /url: https://platform.claude.com/docs
        - listitem:
          - link "Pricing" [ref=e70] [nth=1]:
            - /url: https://claude.com/pricing#api
        - listitem:
          - link "Regional Compliance" [ref=e71]:
            - /url: https://claude.com/regional-compliance
        - listitem:
          - link "Amazon Bedrock" [ref=e72]:
            - /url: https://claude.com/partners/amazon-bedrock
        - listitem:
          - link "Google Cloud’s Vertex AI" [ref=e73]:
            - /url: https://claude.com/partners/google-cloud-vertex-ai
        - listitem:
          - link "Console login" [ref=e74]:
            - /url: https://platform.claude.com/
      - heading "Learn" [ref=e75] [level=3]
        - listitem:
          - link "Blog" [ref=e76]:
            - /url: https://claude.com/blog
        - listitem:
          - link "Claude partner network" [ref=e77]:
            - /url: https://claude.com/partners
        - listitem:
          - link "Connectors" [ref=e78]:
            - /url: https://claude.com/connectors
        - listitem:
          - link "Courses" [ref=e79]:
            - /url: /learn
        - listitem:
          - link "Customer stories" [ref=e80]:
            - /url: https://claude.com/customers
        - listitem:
          - link "Engineering at Anthropic" [ref=e81]:
            - /url: /engineering
        - listitem:
          - link "Events" [ref=e82]:
            - /url: /events
        - listitem:
          - link "Powered by Claude" [ref=e83]:
            - /url: https://claude.com/partners/powered-by-claude
        - listitem:
          - link "Service partners" [ref=e84]:
            - /url: https://claude.com/partners/services
        - listitem:
          - link "Startups program" [ref=e85]:
            - /url: https://claude.com/programs/startups
        - listitem:
          - link "Tutorials" [ref=e86]:
            - /url: https://claude.com/resources/tutorials
        - listitem:
          - link "Use cases" [ref=e87]:
            - /url: https://claude.com/resources/use-cases
      - heading "Company" [ref=e88] [level=3]
        - listitem:
          - link "Anthropic" [ref=e89]:
            - /url: /company
        - listitem:
          - link "Careers" [ref=e90]:
            - /url: /careers
        - listitem:
          - link "Economic Futures" [ref=e91] [nth=1]:
            - /url: /economic-index
        - listitem:
          - link "Research" [ref=e92] [nth=1]:
            - /url: /research
        - listitem:
          - link "News" [ref=e93] [nth=1]:
            - /url: /news
        - listitem:
          - link "Responsible Scaling Policy" [ref=e94]:
            - /url: https://www.anthropic.com/news/announcing-our-updated-responsible-scaling-policy
        - listitem:
          - link "Security and compliance" [ref=e95]:
            - /url: https://trust.anthropic.com/
        - listitem:
          - link "Transparency" [ref=e96]:
            - /url: /transparency
      - heading "Help and security" [ref=e97] [level=3]
        - listitem:
          - link "Availability" [ref=e98]:
            - /url: https://www.anthropic.com/supported-countries
        - listitem:
          - link "Status" [ref=e99]:
            - /url: https://status.anthropic.com/
        - listitem:
          - link "Support center" [ref=e100]:
            - /url: https://support.claude.com/en/
      - heading "Terms and policies" [ref=e101] [level=3]
        - button "Privacy choices" [ref=e102]
        - listitem:
          - link "Privacy policy" [ref=e103]:
            - /url: https://www.anthropic.com/legal/privacy
        - listitem:
          - link "Consumer health data privacy policy" [ref=e104]:
            - /url: https://www.anthropic.com/legal/consumer-health-data-privacy-policy
        - listitem:
          - link "Responsible disclosure policy" [ref=e105]:
            - /url: https://www.anthropic.com/responsible-disclosure-policy
            - /url: https://www.anthropic.com/legal/commercial-terms
            - /url: https://www.anthropic.com/legal/consumer-terms
        - listitem:
          - link "Usage policy" [ref=e106]:
            - /url: https://www.anthropic.com/legal/aup
    - contentinfo: © 2025 Anthropic PBC
    - navigation "Social media links" [ref=e107]:
      - listitem:
        - link "Visit our LinkedIn page" [ref=e108]:
          - /url: https://www.linkedin.com/company/anthropicresearch
      - listitem:
        - link "Visit our X (formerly Twitter) profile" [ref=e109]:
          - /url: https://x.com/AnthropicAI
      - listitem:
        - link "Visit our YouTube channel" [ref=e110]:
          - /url: https://www.youtube.com/@anthropic-ai


agent-browser snapshot -i -c
出力 
- link "Skip to main content" [ref=e1]
- link "Skip to footer" [ref=e2]
- link "Home" [ref=e3]
- link "Research" [ref=e4]
- link "Economic Futures" [ref=e5]
- button "Commitments" [ref=e6]
- button "Learn" [ref=e7]
- link "News" [ref=e8]
- link "Try Claude" [ref=e9]
- link "Start building" [ref=e10]
- link "Developer docs" [ref=e11]
- link "Featured Demystifying evals for AI agents The capabilities that make agents useful also make them difficult to evaluate. The strategies that work across deployments combine techniques to match the complexity of the systems they measure." [ref=e12]
- link "Effective harnesses for long-running agents Nov 26, 2025" [ref=e13]
- link "Introducing advanced tool use on the Claude Developer Platform Nov 24, 2025" [ref=e14]
- link "Equipping agents for the real world with Agent Skills Oct 16, 2025" [ref=e15]
- link "Building agents with the Claude Agent SDK Sep 29, 2025" [ref=e16]
- link "Effective context engineering for AI agents Sep 29, 2025" [ref=e17]
- link "A postmortem of three recent issues Sep 17, 2025" [ref=e18]
- link "Writing effective tools for agents — with agents Sep 11, 2025" [ref=e19]
- link "How we built our multi-agent research system Jun 13, 2025" [ref=e20]
- link "Raising the bar on SWE-bench Verified with Claude 3.5 Sonnet Jan 06, 2025" [ref=e21]
- link "Building effective agents Dec 19, 2024" [ref=e22]
- link "Introducing Contextual Retrieval Sep 19, 2024" [ref=e23]
- link "See open roles" [ref=e24]
- link "Return to homepage" [ref=e25]
- link "Claude" [ref=e26]
- link "Claude Code" [ref=e27]
- link "Claude in Chrome" [ref=e28]
- link "Claude in Excel" [ref=e29]
- link "Claude in Slack" [ref=e30]
- link "Skills" [ref=e31]
- link "Max plan" [ref=e32]
- link "Team plan" [ref=e33]
- link "Enterprise plan" [ref=e34]
- link "Download app" [ref=e35]
- link "Pricing" [ref=e36]
- link "Log in to Claude" [ref=e37]
- link "Opus" [ref=e38]
- link "Sonnet" [ref=e39]
- link "Haiku" [ref=e40]
- link "AI agents" [ref=e41]
- link "Code modernization" [ref=e42]
- link "Coding" [ref=e43]
- link "Customer support" [ref=e44]
- link "Education" [ref=e45]
- link "Financial services" [ref=e46]
- link "Government" [ref=e47]
- link "Healthcare" [ref=e48]
- link "Life sciences" [ref=e49]
- link "Nonprofits" [ref=e50]
- link "Overview" [ref=e51]
- link "Developer docs" [ref=e52] [nth=1]
- link "Pricing" [ref=e53] [nth=1]
- link "Regional Compliance" [ref=e54]
- link "Amazon Bedrock" [ref=e55]
- link "Google Cloud’s Vertex AI" [ref=e56]
- link "Console login" [ref=e57]
- link "Blog" [ref=e58]
- link "Claude partner network" [ref=e59]
- link "Connectors" [ref=e60]
- link "Courses" [ref=e61]
- link "Customer stories" [ref=e62]
- link "Engineering at Anthropic" [ref=e63]
- link "Events" [ref=e64]
- link "Powered by Claude" [ref=e65]
- link "Service partners" [ref=e66]
- link "Startups program" [ref=e67]
- link "Tutorials" [ref=e68]
- link "Use cases" [ref=e69]
- link "Anthropic" [ref=e70]
- link "Careers" [ref=e71]
- link "Economic Futures" [ref=e72] [nth=1]
- link "Research" [ref=e73] [nth=1]
- link "News" [ref=e74] [nth=1]
- link "Responsible Scaling Policy" [ref=e75]
- link "Security and compliance" [ref=e76]
- link "Transparency" [ref=e77]
- link "Availability" [ref=e78]
- link "Status" [ref=e79]
- link "Support center" [ref=e80]
- button "Privacy choices" [ref=e81]
- link "Privacy policy" [ref=e82]
- link "Consumer health data privacy policy" [ref=e83]
- link "Responsible disclosure policy" [ref=e84]
- link "Usage policy" [ref=e85]
- link "Visit our LinkedIn page" [ref=e86]
- link "Visit our X (formerly Twitter) profile" [ref=e87]
- link "Visit our YouTube channel" [ref=e88]

AIエージェントに使わせる方法

プロンプト

cliをインストールした状態で普通にプロンプトで指示する

agent-browserを使用して最新のLLM論文の情報を収集してください。 利用可能なコマンドを確認するには、`agent-browser --help` を実行してください。

AGENTS.md / CLAUDE.md

AGENTS.md / CLAUDE.mdに直接書いてしまうのもあり

公式の例

## Browser Automation

Use `agent-browser` for web automation. Run `agent-browser --help` for all commands.

Core workflow:
1. `agent-browser open <url>` - Navigate to page
2. `agent-browser snapshot -i` - Get interactive elements with refs (@e1, @e2)
3. `agent-browser click @e1` / `fill @e2 "text"` - Interact using refs
4. Re-snapshot after page changes

Claude Code Skill / Agent Skills

リポジトリのagent-browser/skills/agent-browserにskillがあります。
必要なタイミングで読みにいくのでコンテキストの節約になります。

macOS/Linux用コマンド

mkdir -p .copilot/skills/agent-browser
curl -o .copilot/skills/agent-browser/SKILL.md \
  https://raw.githubusercontent.com/vercel-labs/agent-browser/main/skills/agent-browser/SKILL.md

mkdir -p .claude/skills/agent-browser
curl -o .claude/skills/agent-browser/SKILL.md \
  https://raw.githubusercontent.com/vercel-labs/agent-browser/main/skills/agent-browser/SKILL.md

Windows (コマンドプロンプト) 用コマンド

mkdir .copilot\skills\agent-browser
curl -o .copilot\skills\agent-browser\SKILL.md https://raw.githubusercontent.com/vercel-labs/agent-browser/main/skills/agent-browser/SKILL.md

mkdir .claude\skills\agent-browser
curl -o .claude\skills\agent-browser\SKILL.md https://raw.githubusercontent.com/vercel-labs/agent-browser/main/skills/agent-browser/SKILL.md

Windows (PowerShell) 用コマンド

New-Item -ItemType Directory -Force -Path .copilot/skills/agent-browser
curl -o .copilot/skills/agent-browser/SKILL.md https://raw.githubusercontent.com/vercel-labs/agent-browser/main/skills/agent-browser/SKILL.md

New-Item -ItemType Directory -Force -Path .claude/skills/agent-browser
curl -o .claude/skills/agent-browser/SKILL.md https://raw.githubusercontent.com/vercel-labs/agent-browser/main/skills/agent-browser/SKILL.md

参考

8
4
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
Sign upLogin
8
4

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?