Five copy-paste prompts that turn watch output into an artifact. Pick
the one that matches your goal, paste it above the watch block, hand
the whole thing to your agent.
Each prompt assumes you ran watch <url> and now have:
VIDEO: <path>
DURATION: <seconds>
FRAMES:
<path1>
<path2>
…
TRANSCRIPT:
<full text>
| File | When to use |
|---|---|
| implement-from-video.md | Tutorial / coding walkthrough → working code |
| extract-architecture.md | System / architecture talk → interactive diagram |
| clone-ux.md | UI / motion demo → working React component |
| paper-to-code.md | ML paper or research talk → runnable notebook |
| tutorial-walkthrough.md | Long tutorial → AI type-along, step by step |
Mix and match: nothing stops you from running the same watch output
through two prompts and comparing artifacts.