REPO PACKAGE
browserbase skills
OFFICIAL16skills
1ready to use
LABSCO SUMMARY
The fourteen automation skills split into a basic layer and a debugging layer. browser, browserbase-cli, functions, cookie-sync, fetch, and search cover the building blocks — opening a remote or local session, running platform API commands, deploying automation as a cloud function, reusing your logged-in cookies, and fetching a URL or a search result without a full browser. On top of that sit browser-trace and browser-to-api, which turn a captured run into a per-page trace or a best-effort OpenAPI spec; autobrowse, which iterates a task against its own trace until it passes; ui-test and agent-experience, which run adversarial QA and developer-experience audits; safe-browser, which builds a domain-allowlisted agent with no raw CDP access; webmcp-gen, which authors WebMCP init scripts; and browser-use-to-stagehand, which migrates existing browser-use (Python) scripts onto Stagehand v3.
company-research, competitor-analysis, and event-prospecting are a different product: all three call the Browserbase Search API to find companies or people, score them against a stated ICP, and compile the results into an HTML report — a target-company list, a competitor comparison with pricing and mentions, or a conference speaker list ranked by fit. None of the three build or test browser automation; they read as sales and GTM tooling that happens to ship in the same repository.
READ THE FULL ANALYSIS
One skill ships blind. browserbase-cli carries no description in its own frontmatter at all — the README says it wraps the same CLI for sessions, projects, contexts, extensions, and the dashboard, but nothing in the skill file itself tells an agent when to reach for it, so that characterization comes from the README rather than the skill's own trigger text.
What it costs to run any of them is not yet checked. All 17 skills carry an unverified dependency label rather than ready, account_key, or local_tool, so we can't yet say which ones need a Browserbase account and API key versus which run against a local Chrome profile with --local. Given that browse talks to Browserbase's own session infrastructure, assume most of the automation layer needs an account until that gets checked.
Checked from the repository's own skill files and README, 19 September 2026; we have not run any of the 17 end to end.
ALSO IN THIS PACKAGE
PLUGINS
6 itemsPLUGIN
browse
Automate web browser interactions using natural language. Use when the user asks to browse websites, navigate web pages, extract data from websites, take screenshots, fill forms, click buttons, or interact with web applications.
PLUGIN
functions
Deploy serverless browser automation to the cloud using Browserbase Functions. Use when the user wants to deploy automation to run on a schedule, needs a webhook endpoint for browser automation, wants to run automation in the cloud instead of locally, or asks about Browserbase Functions.
PLUGIN
browser-trace
Capture a full DevTools-protocol trace (CDP firehose, screenshots, DOM dumps) alongside any browser automation, then bisect the stream into per-page searchable buckets. Use when the user wants to debug a failed run, audit network/console/DOM activity, attach a trace to an in-flight Browserbase session, or feed structured per-page summaries back into an agent loop.
PLUGIN
safe-browser
Build local constrained-browser agents with a safe_browser tool that owns CDP, enforces a domain allowlist with Fetch interception, and lets a runtime Claude Agent SDK agent complete browsing tasks without raw browser, shell, or CDP access.
PLUGIN
webmcp-gen
Author, compile, and validate site-specific WebMCP init scripts from a target URL and desired tool capability.
PLUGIN
add-webmcp
Analyze an existing web application, add first-party WebMCP tools backed by its routes, forms, actions, and schemas, and validate them with Stagehand v4.
PROMPT TEMPLATES
2 itemsPROMPT TEMPLATE
playwright
PROMPT TEMPLATE
stagehand
WHAT'S INSIDE
16 showing · 16 totalReady to use
Nothing else to set up — install it and go.
16 of 16
SETUP
Showing all 16 skills
agent-experience
Sets several AI assistants loose on your product's documentation with nothing but a one-line task each, then grades how far they got and what tripped them up.
autobrowse
Point it at a job on a website — book a flight, fill in a form — and it keeps retrying and correcting its own instructions until that job works every time.
browser
Drives a real web browser — opening pages, clicking, filling forms, reading the page — on your own computer or a cloud one that gets past CAPTCHAs and bot blocks.
browser-to-api
A site you want to pull data from has no published API — this reads the requests its own pages make and reconstructs one, as documentation plus working sample code.
browser-trace
A flight recorder for a browser being driven by software: it watches without interfering, saving every request, error and screenshot so you can replay why a run went wrong.
browser-use-to-stagehand
Rewrites a browser-automation script from browser-use in Python into Stagehand in TypeScript, deciding step by step which parts genuinely need AI and which can simply be spelled out.
company-research
You give it your own company's name; it goes and finds businesses that would plausibly buy from you, researches each one, and hands back a scored shortlist.
competitor-analysis
A competitor research job run end to end: it works out which companies you are actually up against, digs through each one's site, prices, press and user chatter, and hands back a browsable report instead of a pile of notes.
cookie-sync
Copies the logins from the Chrome on your own machine up to a cloud browser, so it can open sites as you instead of hitting a sign-in wall.
event-prospecting
Takes a conference's speaker list and narrows it to the people worth contacting — only those whose companies could actually buy from you, each with a reason to get in touch.
fetch
Grabs a page's contents in a single request without starting a browser at all — the fast, cheap option when the page has nothing to click and nothing to render.
functions
Puts a browser-automation script on Browserbase's servers so it runs there instead of on your machine — on a schedule, from a webhook, or whenever you call it.
safe-browser
Builds a working example of an AI agent that can only visit the websites you approve — every other address is blocked by the browser itself, not by telling the agent to behave.
search
Runs a web search and hands back the results as data — links, titles, authors, dates — instead of a page you would have to read and click through.
ui-test
Opens your app in a real browser and deliberately tries to break it, instead of just confirming it works. It can test only what a recent code change touched, or roam the whole app hunting for bugs nobody thought to look for.
webmcp-gen
Gives a website a short list of actions an AI can call straight out — read the page's details, run its search — instead of making it click through the interface, then checks in a real Chrome that each one registers and returns what it should.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (16 skills)
npx skills add browserbase/skillsSINGLE SKILLOr install a single skill
npx skills add browserbase/skills --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.
PLUGIN MARKETPLACEAdd the marketplace, then install the plugins you want
/plugin marketplace add browserbase/skills/plugin install browse@browserbase/plugin install functions@browserbase/plugin install browser-trace@browserbaseTyped inside the agent's own prompt, not in a terminal. The marketplace lists 6 plugins; three are shown. Every one installs the same way, with its own name before @browserbase.