Most browser agents click around websites like a user. Quark goes deeper. It sits passively in your browser, maps every API call a site makes to its own backend, then lets you describe what you want in plain English.

It generates a script that modifies the site permanently using that live API context.

In the demo I searched Amazon and typed one sentence. It built a floating CSV exporter using Amazon's own internal search API. No HTML scraping, no manual selectors. Same approach works on LinkedIn, GitHub, Jira, HubSpot, or any site with network traffic.

Scripts persist per domain and re-inject on every page load. It is less like a task agent and more like a permanent layer on top of any website you use.

GitHub: https://github.com/hvardhan878/quark-browser-agent Would love feedback on the architecture and whether this approach holds up long term given Chrome MV3.