@53able/pi-agent-browser
A browser automation toolkit within the Pi framework
Install & Use
Copy this prompt and send it to your AI assistant (Claude / Cursor / TRAE / Codex / WorkBuddy etc.) to auto-install:
Help me install this AI Skill: @53able/pi-agent-browser. It is used for: A browser automation toolkit within the Pi framework Full Skill content: https://321skill.com/skills/53able-pi-agent-browser/raw/index.md Read that page and install it.
The prompt includes a link to the full Skill content. You can also view the full content.
This Skill addresses the issue where Pi agents, when needing to operate real web pages, extract rendered content, maintain login states, or take screenshots for evidence, rely on ad-hoc shell commands that are fragile and uncontrollable. In practical development, developers often require AI agents to automatically browse web pages, fill out forms, and scrape data, but lack type-safe and reusable tools.
Usage is straightforward: simply install agent-browser and Chrome for Testing, then install this package via pi install. After that, you can invoke over 10 tools using natural language instructions: open a URL, read page text, get an accessibility tree snapshot, click/fill forms, take screenshots, execute JavaScript, manage session states, hand over to human, etc. A typical workflow is: first open a page, then get a snapshot to obtain an @ref, followed by clicking or filling forms, and finally reading or taking screenshots to save evidence.
It is well-suited for teams or individuals who need Pi agents to perform browser automation tasks. Particularly for those already using agent-browser to manage browser sessions, require persistent login states (e.g., maintaining cookies), or need to restrict access domains for security. Test engineers can use it for end-to-end testing, and backend developers can use it to scrape dynamically rendered pages.
It is recommended to consistently use this tool when stable browser interaction is needed, avoiding makeshift shell commands. Note that it primarily relies on agent-browser and Chrome for Testing; you must ensure both are correctly installed and available in the PATH. It is also advised to set allowedDomains to limit the access scope and prevent the agent from mistakenly operating on external sites.
Key Features
Unlike using the `agent-browser` command line directly, this Skill provides independent, typed Pi tools for each browser operation (e.g., `agent_browser_click`, `agent_browser_fill`), instead of a single free-form string command, thereby offering better type safety, parameter validation, and composability.
Limitations
Depends on `agent-browser` and Chrome for Testing. Ensure `agent-browser` is in the PATH or set the `AGENT_BROWSER_BIN` environment variable. Only supports the Pi/Feynman runtime.
FAQ
How do I install this Skill?
Run `pi install npm:@53able/pi-agent-browser`, or install from GitHub: `pi install https://github.com/53able/pi-agent-browser`.
How can I reuse login states across sessions?
Use the `session` parameter to specify a name and set `restore: true`. The state will be automatically saved after the first login and can be reused by opening the same session subsequently.
Installation guide for AI assistants
If your AI coding assistant (Claude Code, Cursor, TRAE etc.) can see this page, send it this message to auto-install:
Visit https://321skill.com/skills/53able-pi-agent-browser/raw/index.md to read the original Skill definition (Markdown format) for @53able/pi-agent-browser, and install it according to the instructions.
Raw Markdown URL for AI: /skills/53able-pi-agent-browser/raw/index.md