watch-videolisted

# watch-video Use this skill when a user asks you to analyze a video URL, local video, screen recording, tutorial, demo, UI bug recording, product walkthrough, or any task where visible UI/actions and spoken content matter. ## Locating The Scripts Every command below runs a script under this skill's `scripts/` directory, which sits next to this `SKILL.md`. Resolve that skill directory and run scripts relative to it; do not rely on a client-specific environment variable. Supported platforms are macOS and Linux. ## Operating Rules - Ask the user which detail level to run for each new video (see "Ask The Detail Level First"); skip the question only when the request or environment already answers it. - Prefer native captions/transcripts when available. For URLs the script probes captions first and skips the media download when usable captions cover transcript detail and no cue timestamps are pinned. - Use `yt-dlp` for URL metadata, captions, and media; `ffmpeg`/`ffprobe` for audio clips and frames. - Use Groq Whisper as the default fallback when captions are missing or obviously incomplete. Audio beyond the 24 MiB safety threshold, which is below providers' 25 MB upload limit, is chunked and stitched automatically; partial chunk failures degrade to a partial transcript instead of failing the run. - Use OpenAI transcription only when explicitly requested with `--transcriber openai`. - Default Groq model: `whisper-large-v3-turbo`. Default OpenAI model: `w