Browser Ops Pro

Checking the same admin pages and dashboards by hand is slow, and letting an AI click around a logged-in site is risky. Browser Ops Pro gives your AI a safer middle path: it writes a short JSON plan, checks it, runs it with Playwright and the Chrome or Edge you already have, and leaves a report you can read.

Plans use 23 actions to navigate, wait, assert, extract text, attributes, tables and counts, take screenshots, hover, scroll, and (only when allowed) click, fill and submit. A validate mode catches mistakes without opening a browser. An allowedOrigins list blocks navigation to any other site before the request leaves the browser. Changing steps need allowMutations, and a step marked confirm stops the run with a screenshot until you rerun with --yes. Secrets come from environment variables and are redacted from every output, and filled secret fields are masked in screenshots.

Each run writes result.json, a Markdown report, failure screenshots, per-step timing and retry counts, and can keep a JSONL history that shows what changed since the last run. A login bootstrap lets you log in by hand once into a saved profile, and a session check tells you whether that login still works.

Inside: three Node.js scripts with --help and an offline --self-test, a plan schema reference, workflow and browser setup guides, a demo page with example plans, and a test log. Old 1.x plans still run.

You install playwright-core yourself (no browser download). Chromium-based browsers only. It doesn't solve captchas or 2FA, and it isn't a large-scale scraper. Saved profiles hold live logins, so keep them private.