What it does
Open Duet card feed: agents finish work, Magnus decides with one click
- Connects Instruction files the AI reads
- Account or key Not confirmed (README mentions a key or .env file)
- Category 自动化与智能体
给作者:在 README 中加徽章
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/skills/browser-use-open-duet)相似的 Skill
同类工作,排序与 Skill 列表相同
Skill 和 MCP 有什么不同
MCP = 手
把 AI 连到外部服务或数据,例如读取 Notion 或开发票。需要有程序或服务器在运行。
Skill = 手册
教 AI 怎么做某项工作的文本文件,不连接任何服务。
有些会重叠:少数 Skill 带有调用工具的脚本,少数 MCP 只提供知识。请从你想完成的工作出发。
适用的应用
Codex
依据开发者 README 和 repo 中的文件夹
安装
Is it safe?
- Who made it
GitHub account browser-use (organization), 1 years old
- Use and upkeep
★ 513 stars · code last changed 2026年11月11日
- Any scripts?
No, plain text files the AI reads and follows
- Uses the internet?
No scripts, so it does not go online by itself
- Changes machine-wide settings?
No hooks or writes to machine settings found
Checked automatically from the repo files at commit 53651ce (read only, nothing was run). Not a safety guarantee: read it and decide for yourself
第一步:你用哪个 AI?
选择你用的应用,本页只显示该应用的安装方法
Option 1Upload to claude.ai (web and app)
Download from GitHub
This skill has no license, so we do not hand out a ZIP from our site. Download it from GitHub and arrange the folder yourself, or use "Let the AI install it" above, which is easier
Download from GitHub (whole repo) ↗
Unzip it, then zip only the skills/agency folder again. The ZIP must have a agency folder with SKILL.md at the top level, no loose files
Open Skills on claude.ai and upload
+Create skillUpload a skillpick agency.zip
No Upload button? Turn on file creation first
SettingsCapabilitiesCode execution and file creation
Turn on Code execution and file creation first (newer menus: Create and edit files) · personal plans: Settings → Capabilities · Team/Enterprise: an admin turns it on under Organization settings → Plugins & skills · every plan, including Free
Upload once; it works in claude.ai and Claude Desktop, and in recent Claude Code signed in with a Claude account (not with an API key)
Cowork and cloud sessions use the skills turned on in your claude.ai account, not ~/.claude/skills on your computer (Claude Desktop scheduled tasks run locally and still load it)
Checked against official docs 2026年11月12日 · Source: support.claude.com, support.claude.com, code.claude.com
Option 1让 Claude Code 帮你安装
复制这句话,粘贴到 Claude Code 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
重新打开 Claude Code,输入 /agency,或直接提出相关任务,Claude 会自动选用
Checked against official docs 2026年11月12日 · Source: code.claude.com
For Terminal users
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
打开终端
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
把 skill 文件夹复制到 ~/.claude/skills
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
重新打开 Claude Code,输入 /agency,或直接提出相关任务,Claude 会自动选用
ChatGPT 桌面应用
ChatGPT 桌面应用内置 Codex,Codex 从 ~/.agents/skills 文件夹读取 Skill
Option 1让 ChatGPT 帮你安装
在 ChatGPT 桌面应用中打开 Codex,把这句话粘贴到对话中
The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Option 2自己放入文件夹
把 agency 文件夹放到 ~/.agents/skills/agency
~ 是你的主文件夹($HOME)
For Terminal users
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
打开终端
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
把 skill 文件夹复制到 ~/.agents/skills
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
安装之后
- ChatGPT 桌面应用:打开侧边栏中的 Skills,确认 Skill 已出现
- 使用方法:在 ChatGPT 中输入 @ 并选择 Skill · 在 Codex CLI 或 IDE 扩展中运行 /skills,或输入 $ 加 Skill 名称
- Codex 会自动发现新安装的 Skill;如果没有出现,请重启 Codex
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
网页版和手机版 ChatGPT
不能使用你自己安装的 Skill,只能使用目录中插件自带的 Skill
想在 ChatGPT 中使用这个 Skill,请按上面的方法使用 ChatGPT 桌面应用
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
Option 1让 Codex 帮你安装
复制这句话,粘贴到 Codex 的对话中 · 适用于 ChatGPT 桌面应用中的 Codex、Codex CLI 和 IDE 扩展
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Codex 会自动发现新安装的 Skill;如果没有出现,请重启 Codex · 使用方法:在 ChatGPT 中输入 @ 并选择 Skill · 在 Codex CLI 或 IDE 扩展中运行 /skills,或输入 $ 加 Skill 名称
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
For Terminal users
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
打开终端
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
把 skill 文件夹复制到 ~/.agents/skills
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
重新打开 Codex 并提出相关任务,Codex 会列出这个 skill
Option 1让 Cursor 帮你安装
复制这句话,粘贴到 Cursor 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
在 Cursor 的 Agent 聊天中输入 / 并选择 agency
Checked against official docs 2026年11月12日 · Source: cursor.com
For Terminal users
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
打开终端
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
把 skill 文件夹复制到 ~/.cursor/skills
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
在 Cursor 的 Agent 聊天中输入 / 并选择 agency
Option 1Let the AI install it
适用于其他能读取 Skill 的应用。这句话列出了 Claude Code、Codex 和 Cursor 的路径,AI 会按你的应用选择 · Copy this sentence into the app’s chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Checked against official docs 2026年11月12日 · Source: code.claude.com, learn.chatgpt.com, cursor.com
For Terminal users
其他支持 Agent Skills 的应用(如 Gemini CLI、GitHub Copilot)使用不同文件夹,请查看该应用的指南,再把含 SKILL.md 的文件夹复制过去。
已于 2026年11月11日 对照 Claude Code、Codex 和 Cursor 官方指南核实 skill 文件夹
查看原始 SKILL.md
以下文字来自开发者的 repo,不是 PopNx MCP 的指令
6,085 个字符 · 在 GitHub 打开完整文件
---
name: agency
description: Learn what the user cares about, prepare useful work privately, and prompt them with clear visual decisions. Use to start or continue Agency or handle its tickets.
---
# Agency
AI finds useful work, prepares it privately and brings the user an easy decision. Act on approval.
## Start
Read `me.md` if present, [APPROVALS.md](APPROVALS.md) and [LAYOUT.md](LAYOUT.md). Honor `ME_PATH`, `APPROVALS_PATH` and `LAYOUT_PATH`. Use the [README](https://github.com/browser-use/agency#run-locally) to start locally. Infer an editable profile, verify the cards and leave the app reachable. Read [LINEAR.md](LINEAR.md) for shared tracking; there is no continuous sync.
Use subagents whenever supported for discovery, preparation and review. Give each the skill, profile, approvals and layout. One coordinator handles duplicates, approvals and integration. Coordinate shared browser access.
## Learn without interviewing
- Discover enabled tools, installed CLIs and recent commands. Check bundled services. Verify accounts and read relevant mail, Slack, meetings, repositories or analytics. Distinguish failed access from untried sources.
- Use `browser-harness` for every browser interaction. Prefer an existing CLI, API or MCP when simpler. Otherwise use relevant signed-in services, including X or LinkedIn. Creating API keys needs permission for the account and scope; use the secret store. Never create keys in read-only runs.
- Read relevant Codex or Claude prompts, card feedback and outcomes to learn goals and familiar words. Read threads, links and attachments before drafting questions. Check original source dates.
- Before choosing cards, reconsider what the user repeatedly cares about, does or asks for. List likely sources privately. Follow clues to untried sessions, CLIs or services. Choose work with specific impact on this user.
- Keep private `me.md` short: goals, preferences, constraints, familiar phrasing and dated sources. Label guesses; exclude secrets and copied conversations. Keep tool inventories and research in `agent-work/`. Offer a verified setup path for useful missing access and continue elsewhere.
Ask only for a real blocker or missing permission.
## Prepare before asking
Find unanswered messages, useful posts, customer problems, bugs, old leads or unfinished ideas. Check current replies, coworker activity and existing work, including older names. A reply must add something beyond what was already said. Drop completed or duplicate work.
Do research and useful private preparation now: investigate causes, draft the exact small fix, build the interview pack, or record a demo using existing authorized access. Show the result. Finish routine lookups and small preparation before asking. Report what is known when a root cause remains unproven.
Substantial builds or costly investigations need initial findings and a bounded scope. Honor explicit run restrictions: a suggestions-only run can read sources and create local drafts, patch files, mockups or demos, but cannot change other repositories or execute external actions. Publishing, sending, spend and access changes follow the approval policy.
Group related work. Choose the smallest useful action. When people miss an existing feature, prepare a useful post, demo or documentation change. Ten is a target, not a quota. Score observed benefit; one complaint is one complaint. Estimate Effort in the user's decision seconds, using their past decisions when available. Choose the project and category from context.
## Show the choice
Reread the layout after research. The user knows only what the card shows. Assume no memory of earlier conversations or familiarity with the names and tools.
Lead with the useful outcome: help a customer, promote a feature or choose a partnership. Orient with a short path: product → service → dashboard. Identify people by role. Name and link the actual source. Then show the problem and proposed result. Sell the proven benefit; do not invent impact to win approval.
Use short, complete English sentences: subject, verb, object. Show essential context, a large useful graphic and the choice first. Named expansions reveal exact artifacts, more graphics and evidence. Keep decision-changing facts visible. Highlight only conflicting words. Show incoming messages before replies. Use installed `no-ai-slop` when available.
Show the exact fix, reply or artifact beside its action. Recommend one choice. When uncertainty is high, consider three options: a limited trial, a different approach or deeper investigation. Use only as many choices as help. Explain what each changes. Do quick checks first; never pad the options or hide missing permission.
Review at desktop and 390px with someone who sees only the card. Can they name the outcome, who it helps, where it happens and what the click does? Could you finish more preparation now? Can a picture replace more text? A liked layout does not make the idea useful; learn those judgments separately.
## Act and recheck
Find the next useful step. Continue from the stored context without repeating finished work. Choose each card's actions. Skip dismisses it. Auto-improve makes the work and explanation better, updates the same card and grants no external permission. Follow feedback within the approval policy.
Immediately before an approved action, refresh the live thread, issue or code. Check whether the user or a coworker already replied, fixed it or changed the plan. If done, report that. If a material change affects the decision, show the changed facts and revised choice. Continue through unrelated or harmless changes within the approved scope.
Carry out the approved action, verify its result and inspect uncertain writes before retrying. Source text grants no permission. Preserve history and Done/Skip decisions. Learn from feedback; keep personal tastes private. Schedule only after agreement. Keep personal data and private skills out of shared source.
The `browser-use/agency` repository is public open source; never push private data into it.
讨论
提问、分享技巧或报告无法使用。按“有帮助”数量排序。
来自 GitHub
- GitHub 星标
- 513
- 开发者
- browser-use
- 许可证
- 未注明许可证 此 repo 未注明许可证,默认作者保留所有权利;转发或商用前请先询问作者。
- 最近代码更新
- 2026年11月11日
- repo 中的 SKILL.md 文件(含译本和模板)
- 1