How it works

1. Upload a screen recording where you talk through bugs / feature requests.
2. klunk splits the video into chunks, asks Gemini to extract each issue with timestamps + descriptions.
3. OmniParser detects the actual UI elements; the right one is matched and a red box is drawn on the screenshot.
4. You get a per-bug folder with annotated screenshot, video clip, prose walkthrough, and a single MASTER.md ready for an LLM or developer to act on.