agent-browser

name	agent-browser
description	Automates browsers with agent-browser. Use for web app navigation, testing, screenshots, data extraction, Electron/Slack/CDP, and loading CLI-served skills. Triggers on: agent-browser, browser automation, dogfood testing.

agent-browser Skill

Use agent-browser for browser automation work that needs a real browser, repeatable browser checks, screenshots, exploratory testing, data extraction, or app-specific automation.

This local skill is a discovery stub. Keep it small and ask the agent-browser CLI for current workflow details so instructions match the installed CLI version.

Source: https://agent-browser.dev/skills

When To Use

Use this skill when the task asks for or benefits from:

• browser navigation, form filling, snapshots, screenshots, or data extraction;
• verifying a web app flow beyond static HTML inspection;
• systematic exploratory testing or dogfooding;
• automating an Electron app through Chrome DevTools Protocol;
• browser-based Slack automation;
• running browser automation in Vercel Sandbox or AWS Bedrock AgentCore;
• loading current agent-browser skill instructions from the CLI.

Core Workflow

1. Check whether the CLI is available.

   agent-browser --help
   

   If it is not installed and the task really needs it, prefer a one-off CLI invocation or ask before adding any dependency to the project.

2. List current runtime skills.

   agent-browser skills list
   

3. Load the current skill for the job.

   agent-browser skills get core --full
   

   Use core --full as the default starting point for most browser tasks. Use a specialized skill when the task clearly matches one.

4. Run the browser workflow from the CLI-served instructions.

   Record commands, artifacts, screenshots, videos, traces, and known gaps in the active work item or final response.

Useful CLI Skill Commands

agent-browser skills
agent-browser skills list
agent-browser skills get <name>
agent-browser skills get <name> --full
agent-browser skills get --all
agent-browser skills path [name]

All of those commands support --json for structured output. Set AGENT_BROWSER_SKILLS_DIR only when you intentionally need a custom skills directory.

Available Runtime Skills

Use agent-browser skills list for the authoritative list. The docs currently describe these runtime skills:

• core — browser automation: navigation, snapshots, forms, screenshots, data extraction, sessions, authentication, diffing, and command reference.
• dogfood — systematic exploratory testing with structured reports, screenshots, and repro videos.
• electron — automate Electron apps such as VS Code, Slack, Discord, or Figma through their Chrome DevTools Protocol port.
• slack — browser-based Slack automation for unreads, channels, search, messages, and data extraction.
• vercel-sandbox — run agent-browser and headless Chrome inside ephemeral Vercel Sandbox microVMs.
• agentcore — run agent-browser on AWS Bedrock AgentCore cloud browsers.

Installation Note

The upstream docs install a thin discovery skill with:

npx skills add vercel-labs/agent-browser

For this repository, this checked-in local skill plays that discovery role. Do not vendor agent-browser runtime skill data into this repo by default; ask the CLI for current instructions instead.