What browser-tests does
Run $browser-tests in Codex or /browser-tests in Claude Code to add or normalize a repeatable browser test harness owned by the project.
The skill reuses an existing browser runner when one is already configured. For compatible JavaScript or TypeScript web and browser-extension projects without a runner, it defaults to Playwright. It adds one small smoke flow, records the exact command as Browser tests in AGENTS.md, and proves the command works.
How Blueprint uses the harness
Once the command is documented:
/featurecan require focused browser coverage for stable behavioral acceptance criteria./implementcan add that focused coverage while building the feature./checkruns the declared browser command and still directly observes claims the harness does not cover./continuousreuses the same command when its Check gate runs.
Continuous Mode does not install browser tooling during a feature. Run /browser-tests before Continuous Mode when the project needs a repeatable browser gate.
What stays optional
Blueprint does not install browser tooling by default. The browser command also stays outside the default Verify recipe and GitHub Actions workflow unless you separately choose to add it.
The skill creates a trustworthy foundation, not a broad end-to-end suite. Each feature should add only the focused coverage needed for its stable behavior.
Browser boundaries
Automated browser tests are strongest for repeatable page flows, forms, navigation, console errors, failed requests, screenshots, and traces. They do not automatically prove authenticated real-profile behavior, browser chrome, extension toolbar interactions, or visual fidelity.
For browser extensions, Playwright can load the built extension in a persistent Chromium context and test its pages and service worker. Opening the real toolbar or side panel may still need a manual browser check.
See Testing and CI, Check, and Continuous for the full verification path.