Describe the request
Chat currently handles screenshots and text well, but local/web videos are a weak point. For game feel work (fishing rod bend, fight timing, camera, SFX), a 2-minute reference clip is more useful than 20 stills.
Today the agent mostly gets page titles, filenames, or a few user-picked frames. It cannot reliably:
watch the clip end to end
report what happens at 0:12 / 0:41 / 1:05
track a moving object (rod blank, line, float)
use audio (reel clicks, rod load, splash)
Please add first-class video ingest in Agent:
Drop a local .mp4 or paste a public URL (YouTube / TikTok / Instagram Reel).
Extract a timeline: keyframes + short motion notes + optional audio events.
Let the user mark ranges (“only 0:18–0:35, the hook-set”).
Return timestamped observations the agent can turn into code changes.
Keep it local-first / opt-in; don’t upload private clips unless the user agrees.
This is high leverage for animation, VFX, UX timing, and any “make it feel like this clip” task. Stills remain useful for color and layout; video is what you need for motion.