Skip to content
Get packs

agent-browser core guide

Lets an agent drive a real Chrome/Chromium browser: open pages, read them, click, fill forms, and take screenshots.

Source Download skill
Repository
vercel-labs/agent-browser
Stars
45,202
Licence
Apache-2.0
Files
14 files

What it does

Once installed, the agent can control a real web browser through a command-line tool instead of guessing at raw HTML. It opens pages, reads a compact accessibility-tree snapshot with numbered element references (such as @e3), then clicks, fills fields, extracts text, takes screenshots, manages tabs, and saves or restores login state. The guide also tells the agent to use a named, isolated browser session rather than the shared default one, and to close the browser when finished.

Great for

  • Developers automating repetitive web tasks such as form filling, scraping or testing a web app
  • QA testers who want an agent to walk through a site and capture screenshots or PDFs
  • Analysts who need data pulled from pages that have no API
  • People who want an agent to log into a site once and reuse the saved session

Skip it if

  • Tasks that must run on sensitive accounts without review: the agent acts with real logins and can submit forms or make purchases, so a human should confirm high-impact actions
  • Non-browser tasks such as local files or desktop apps: the guide itself says to load a different skill for those
  • Sites you are not authorised to access: browser automation can violate terms of service or laws, and the guide covers only the mechanics

Try asking

  • “Open example.com, take a screenshot and save it as home.png”
  • “Fill in the contact form on this page with my details and tell me what error message appears”
  • “Log into my staging app using the saved auth vault and export the dashboard as a PDF”
  • “Extract the price table from this product page into a text file”
Step-by-step guide: how to use agent-browser core guideInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 14 original files, the Apache-2.0 licence and an attribution note.

    core.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip core.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip core.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Start each task with the named session command so you do not interfere with other agents' browsers
  • Re-take a snapshot after any navigation or page change, because old refs can point to the wrong elements
  • Use the auth vault (auth save / auth login) for passwords instead of putting credentials in scripts or prompts

What's inside

  • references/authentication.md
  • references/commands.md
  • references/profiling.md
  • references/proxy-support.md
  • references/session-management.md
  • references/snapshot-refs.md
  • references/streaming.md
  • references/trust-boundaries.md
  • references/video-recording.md
  • references/webgpu.md
  • SKILL.md
  • templates/authenticated-session.sh
  • templates/capture-workflow.sh
  • templates/form-automation.sh

Redistributed unmodified from vercel-labs/agent-browser under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.