lucianguyen

congrats on the launch. curious how you're handling the visually verify the feature looks good part specifically, since that's usually the hardest bit to make reliable compared to the API/CLI checks. are your agents driving a real browser instance for that, or is it screenshot diffing against some baseline been messing around with phi browser lately for a similar problem, it exposes mcp so an agent can drive an actual chromium session with real state/cookies instead of a clean headless instance, made a noticeable difference in catching stuff that only shows up with a logged in session. curious if you rolled your own browser layer for the preview/QA piece or if you're leaning on something existing under the hood also the custom harness decision is interesting, would love to hear more about what specifically off the shelf solutions were missing for the concurrent hundreds of agents case, sounds like a scaling/isolation problem more than a capability one

Since everything is stored as Markdown, how do you usually handle exporting a finished manuscript? Do most users export to DOCX, EPUB, or something else?