video-ingestlisted

# Video Ingest I cannot watch video. This skill turns a video URL into text I can reason over — title, uploader, duration, description, and the full transcript — so the conversation can actually be about the content of the video. ## Steps 1. **Run the collector.** Pass the URL as-is — yt-dlp handles shortened forms, redirects, query strings, and ~1000 sites: ```sh python3 ~/.claude/skills/video-ingest/ingest.py "<url>" ``` 2. **Flags worth knowing:** - `--brief` — metadata + first 800 chars of transcript. Use this first when the user just wants a quick take; re-run without `--brief` if they want to go deep. - `--whisper` — if the video has no subtitles (common on Instagram reels, some X posts, raw uploads), download the audio and transcribe with local faster-whisper. Slower (CPU, ~1× realtime on `base.en`) but works offline. Skip unless needed — most YouTube content has auto-captions. - `--timestamps` — keep `[hh:mm:ss]` markers in the transcript (subtitles only, ignored on whisper). - `--lang xx` — subtitle language root (default `en`). Use when the source is non-English. - `--json` — machine-readable. Use when piping into another skill or saving structured. - `--no-cache` — bypass the `/tmp/video-ingest-cache/` cache. Default behavior caches per video-id so re-asking is instant. 3. **Read the report.** Sections in order: TITLE, UPLOADER, DURATION, UPLOADED, VIEWS, URL, DESCRIPTION, TRANSCRIPT. The transcript line shows the source (`subtitles`