Browse
Use this skill when the task requires browser-based verification instead of static code reading.
This port was reconciled against garrytan/gstack at commit
94993f74012782fd94416dd44b8314f6363a13a4.
Important Adaptation
This repo ports the browse workflow, not the upstream browser daemon, telemetry, or Playwright binary itself. Before testing, read references/runtime-strategy.md and choose the best documented Codex browser surface for the target:
@Browser/ in-app browser for local dev servers, file-backed previews, and public unauthenticated pages@Chromefor signed-in browser state, browser extensions, or existing Chrome-profile context- Computer Use for desktop GUI flows that browser tools or structured integrations cannot cover
- repo-local Playwright only when the target repo already owns the scripts, fixtures, and dependencies
Core Jobs
- verify a page loads and key elements appear
- test a user flow end to end
- capture before/after evidence for bug reports
- inspect console, network, or page state when the tooling allows it
- check responsive layouts and authenticated flows when possible
Workflow
- Choose the runtime path from
references/runtime-strategy.md. - Translate the desired action using
references/browser-command-map.md. - Test the target page or flow.
- Collect evidence: screenshots, repro steps, state changes, and any blind spots.
Guardrails
- Never claim browser automation happened if you only performed HTTP or static checks.
- If authenticated access is required, coordinate with
setup-browser-cookies,@Chrome, or an app-supported test login path. - Report what was tested, what was blocked, and what remains unverified.
- Do not vendor upstream GStack browser runtime pieces into this repo during a QA task.