agent-browser
vercel-labs/agent-browserA native browser-automation CLI for agents, plus 11 skills that teach the snapshot-and-ref workflow for navigating, filling forms, scraping and testing.
14 skills and skill packs reviewed. Sorted by verdict, then depth — how much real expertise ships with the skill — not by popularity.
A native browser-automation CLI for agents, plus 11 skills that teach the snapshot-and-ref workflow for navigating, filling forms, scraping and testing.
Microsoft's command-line interface for driving a browser through Playwright, plus a skill that teaches coding agents the commands.
A skill that lets an agent write and run its own Playwright scripts for testing, screenshots, login flows and arbitrary browser automation.
| Skill | Verdict | Installs |
|---|---|---|
| agent-browser vercel-labs/agent-browser A native browser-automation CLI for agents, plus 11 skills that teach the snapshot-and-ref workflow for navigating, filling forms, scraping and testing. | Recommended | 1.3M |
| playwright-cli microsoft/playwright-cli Microsoft's command-line interface for driving a browser through Playwright, plus a skill that teaches coding agents the commands. | Recommended | 200K |
| Playwright Skill lackeyjb/playwright-skill A skill that lets an agent write and run its own Playwright scripts for testing, screenshots, login flows and arbitrary browser automation. | Recommended | — |
| squirrelscan Skills squirrelscan/skills Two skills and plugins for squirrelscan's website audit CLI: run 300+ SEO, security and accessibility rules, then fix the findings in code and re-audit. | With caveats | 74K |
| Browser Use browser-use/browser-use Seven skills around the Browser Use browser agent: drive local Chrome over CDP, use the paid cloud API, run QA scoring, or pay by crypto wallet. | With caveats | 98K |
| Chrome DevTools MCP ChromeDevTools/chrome-devtools-mcp Google's Chrome DevTools MCP server and CLI, plus seven skills for debugging performance, memory leaks, cookies and accessibility in a live Chrome. | With caveats | — |
| BrowserAct Skills browser-act/skills A browser-automation CLI skill with stealth browsers, CAPTCHA solving and handoff to a human, plus about 100 ready-made scraper skills for shops and lead sources. | With caveats | 193K |
| Firecrawl Skills firecrawl/skills Firecrawl's official catalog of 37 skills for web search, scraping, crawling and browser interaction, plus API integration and workflow recipes. | With caveats | 247K |
| Google Maps Scraper gosom/google-maps-scraper Agent skill that runs the open-source gosom Google Maps scraper locally in Docker to collect business leads, reviews, emails and contact details. | With caveats | — |
| bb-browser epiral/bb-browser A CLI and MCP server that lets agents drive your real, logged-in Chrome and run site adapters (103 commands, 36 platforms), with two small skills describing the commands. | With caveats | — |
| Apify Agent Skills apify/agent-skills Apify's official skills for scraping with its Actor marketplace and for building, converting and integrating Actors with the apify CLI. | With caveats | — |
| shinkoku (Japanese tax filing) kazukinagata/shinkoku A Japanese-language plugin of 24 skills that walks sole proprietors and employees through kakutei shinkoku: bookkeeping, tax calculation and e-Tax entry. | With caveats | — |
| Firecrawl CLI firecrawl/cli Firecrawl's command-line client and 13 agent skills for web search, scraping, crawling, interacting with pages and parsing files through the paid Firecrawl API. | With caveats | 660K |
| terminal-browser zenbu-labs/terminal-browser A real Chromium browser inside a kitty-graphics terminal, with a verify skill that drives apps headlessly in a pty and records a video of the check. | With caveats | — |
An agent browser skill teaches a coding agent to drive a real browser: open pages, fill forms, take screenshots, scrape data and test your app. We reviewed 14. Three get an install verdict and eleven come with caveats. None is rated clean: a browser can reach your logged-in sessions, so every skill here has some access you should know about.
Most of these skills wrap a command line tool or a library, and the skill text teaches the agent the commands. The tool does the work: it launches Chrome or another browser, runs steps and returns page content. Some, like Playwright Skill, have the agent write and run its own scripts.
The browser’s profile decides how much it can see. An in-memory profile starts empty. Attaching to your running Chrome, importing cookies or saving a state file hands the agent whatever those sessions can do. The Claude Code docs warn about prompt injection from servers that fetch external content, and a browser is the clearest case: any page it opens can contain text aimed at your agent.
Check the profile and cookie behavior first. Then check how permissions are granted: several skills pre-approve a wildcard over their own command, which the skills docs describe as a grant for the turn, not a restriction on other tools. A wildcard over the skill’s own CLI is normal. A wildcard over node or bash is much broader.
Next, look at paid services and terms of service. Stealth browsers and CAPTCHA solving carry account and terms risks. Hosted scrapers bill per result. Finally, look at what you must install: a global npm package, a Python CLI, Chrome for Testing.
Of the 14, 8 are marked “review” and 6 have broad access. See the security page for the labels.
agent-browser command, and it can reuse your real Chrome profile or save cookies to state files that stay plaintext unless you set AGENT_BROWSER_ENCRYPTION_KEY. The CLI binary itself was not reviewed.playwright-cli and npx playwright. By default the profile lives in memory, but the skill documents attaching to running Chrome or Edge and reading cookies. It needs Node 18 or newer.node and npm.npx chrome-devtools-mcp@latest and collects Google usage statistics unless you pass --no-usage-statistics.Be careful with BrowserAct Skills: 103 skills with stealth browsers and CAPTCHA solving, and the real instructions come from a CLI installed from PyPI. It has broad access and terms of service risk.
For research workflows built on these tools, see research skills. For testing and code checks, see code review skills. The full list has all 14, and how we review explains the scores.