Core Idea
The author's design philosophy, what sets this apart from peer projects, and quick notes on the key workflows. Skill names mentioned in the text auto-link to their detail pages.
agent-browser is a fast native Rust CLI providing browser automation for AI agents, designed as a composable toolkit rather than a monolithic framework.
Load core first, then invoke specific skills per task. Use semantic locators and snapshots for stable automation.
Skill Composition
Each skill is designed for a specific kind of caller — some only fire when you name them, others load automatically when the model judges them relevant. The split below summarises both modes at a glance.
These skills only fire when you call them by name. The agent will not load them on its own — they stay inert until invoked.
These skills are loaded automatically by the agent when it judges the context relevant. You usually don't need to call them yourself.
Skill List
agent-browser is a fast Rust CLI for AI agents to automate browsers via Chrome CDP with accessibility snapshots. It handles web navigation, form filling, clicking, screenshots, data extraction, and testing, plus Electron/Slack/cloud browser tasks. Value: reliable, version-matched, agent-agnostic web automation without Playwright/Puppeteer.
Runs agent-browser on AWS Bedrock AgentCore cloud browsers for managed browser automation. Uses AWS credentials via env or CLI, standard agent-browser commands with `-p agentcore`, and supports persistent profiles and Live View. Value: scalable, AWS-hosted browser sessions with reused auth and real-time monitoring.
Core usage guide for the agent-browser CLI, a fast Chrome/Chromium automation tool for AI agents. It teaches the snapshot-and-ref workflow to interact with pages using compact @eN refs, covering navigation, clicking, filling, data extraction, screenshots, tabs, auth, and waits. Value: lets agents automate web tasks reliably in ~200-400 tokens per step instead of parsing raw HTML.
Reverse-engineers a website's internal API by recording browser traffic into a HAR file via agent-browser, then generates a standalone client or CLI that calls endpoints directly. Method: record flows, identify endpoints with jq, extract shapes/auth, generate typed functions, verify live. Value: eliminates repeated browser driving, enables fast direct HTTP automation.
Systematically explore and test web apps to find bugs and UX issues via agent-browser. Produces a structured report with step-by-step screenshots, repro videos, and detailed repro steps for every issue, enabling direct handoff to responsible teams for fixing.
Automate any Electron desktop app (Slack, VS Code, etc.) via agent-browser and Chrome DevTools Protocol. Launch app with remote debugging port, connect, snapshot and interact. Enables desktop UI automation, testing, data extraction and multi-app control without native code.
Use browser automation (agent-browser) to interact with Slack workspaces: check unreads, send messages, search, extract data. Delivers unread reports, sent messages, and evidence screenshots without Slack API or tokens.
Run agent-browser + headless Chrome inside Vercel Sandbox microVMs for browser automation from any Vercel-deployed app. Uses on-demand Linux VMs with Chromium and SDK helpers to run commands like screenshot, snapshot, and form fill. Delivers isolated, persistent, binary-limit-free browser automation with sub-second startup via snapshots.
Installation
Copy and run in your terminal to add vercel-labs/agent-browser to your Agent environment.
npx skills add vercel-labs/agent-browser