Connect a web application

RigorRun drives a real browser to the page a person would open first, and watches your agent work.

A browser cannot check its own work

A page saying “done” is a claim by the same system that would have to be wrong for it not to be done. So a browser connection has a state-read capability of none, every verdict from one says OBSERVATIONAL, and the type refuses to let a browser be its own verifier.

That is the honest ceiling, not a gap to close.

Making it stronger

Attach an MCP server or an OpenAPI document for the same system. Then the clicking is watched in the page while the verdict comes from records, and the run is PARTIAL rather than OBSERVATIONAL. Most systems that look API-less have something that can be read.

Playwright is not bundled

playwright-core is roughly 300 MB once its browsers are installed, and most projects never need one, so it is not a dependency. A tool people try with npx in ten minutes cannot open with that download.

npm i -D playwright-core && npx playwright install chromium

Missing, it produces a sentence rather than a module-not-found.

It stays on the site you named

A page can link anywhere, and following one off-site would mean driving a browser to an address you never named. RigorRun does not leave the origin you gave it.

Recordings are not contracts

rigorrun record writes a record of what happened in a page. rigorrun compile cannot read it: compiling needs the state your system held before and after the job, and a browser recording of an uninstrumented application does not carry that.