The skill

The skill is called browseokay. It's a folder your agent loads: SKILL.md plus references/ (API signatures, install notes, clearing state) and learnings/ (site-specific notes). The agent reads it before its first browser task. The skill isn't public yet; install instructions are coming soon.

What it covers

• Running scripts — the agent sends a JavaScript flow to Browseokay's helper command as a heredoc. Scripts run in Node.js; page-side code goes inside page.evaluate().
• Task spaces and pages — create, resume, adopt and release pages; see Task spaces.
• Page actions — goto, click, fill, selectOption, press, hover, drag and drop, waits, and snapshots.
• Visual pages — screenshot(), mouse and keyboard for canvas-style apps.
• Files — setInputFiles() and file choosers for uploads; download events with saveAs().
• Data — page.evaluate() for bulk extraction, page.fetch() for requests with the page's session.
• User control — handOff() when a human must act, and finish() when the job is done.

Site learnings

Notes under learnings/<site>/ are read on every run. The skill ships with learnings for GitHub, Google and X.

Small, explicit API

The runtime deliberately exposes a small custom API. It looks like Playwright in places but isn't: agents are told to use only the documented methods, with page.evaluate() and page.cdp() as escape hatches.