Hey Vercel Community, and @pawlean 
We just shipped a browser automation CLI for our cloud infra, plus the steel-browser skill now live on skills.sh.
Building on the agent-browser (h/t @ctate) , our humans can now let their agents roam the web and:
-
run multi-step flows (login + dynamic UI)
-
return clean markdown, plus screenshots/PDFs as evidence
-
keep session state across longer runs
-
handle blockers (including stealth / captcha flows)
-
run background browser sessions massively in parallel
Thread: https://x.com/steeldotdev/status/2028855809233526799
- Skill install:
npx skills add steel-dev/cli --skill steel-browser
Would love feedback from folks building agent workflows:
-
where do your agents most often break in the browser?
-
what “skill contract” primitives would you want (outputs, artifacts, retries, etc.)?
Thanks, best,
Niko