Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →watch-together
Watch TV shows and movies with the user in real time by processing screen captures into frames and audio analysis.
Passed 21 of 23 test cases with the skill, and 6 of 23 without it.
and 20 more in the full report
16 fixed7 passed either way2 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 16 test cases the skill fixed. Same model, same prompt, with and without the skill.
When starting a local file watch session, my current chat workspace folder is `/workspace/chats/2024-03-29_191a7dcc-3e4d-4825-a5b6-97876525f56c`. What exact value should I supply as the conversation ID parameter?
| Checked | Without skill | With skill |
|---|---|---|
| Instructs passing only the bare UUID 191a7dcc-3e4d-4825-a5b6-97876525f56c | Missed | Passed |
The instructions your model reads when the skill loads.
Real-time co-watching. A cheap, fast "editor" model watches playback continuously; the assistant is woken only at moments worth reacting to, with the exact frames that carry the story and the verbatim dialogue since their last look. Wakes arrive automatically via signal files — no polling, no manual triggering.
Source mode (preferred) — watch-file.py plays a media file in mpv and
reads the source directly: no screen capture, no loopback audio driver, no
capture encode, and the movie can be fullscreen on any display. Frames are
extracted from the source at full quality, subtitles (sidecar .srt or an
embedded text track) become exact dialogue, timestamps are media time, pause
pauses the flow, and seeks resync. Works for anything mpv can play: local
files, network streams, yt-dlp-supported URLs.
Screen mode (fallback) — capture-live.sh records the screen with
ffmpeg for content that only plays in a browser (DRM streaming). Requires
the BlackHole loopback driver for system audio (brew install blackhole-2ch plus a Multi-Output Device in Audio MIDI Setup); without it,
video-only.
Both modes feed the same editor pipeline.
editor.py): a fast vision model that
watches it (video and audio), flags the frame timestamps that carry
the story, transcribes dialogue when subtitles aren't available, and
decides whether to wake the assistant now or hold while a moment is
still building[WATCH] messagetest cases passed, vs 6 / 23 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.