菜单
Download a video from YouTube, Instagram, TikTok, Twitter/X, Facebook, Vimeo (and 1000+ yt-dlp sites) to local disk, and optionally transcribe it. Standalone utility for repurposing and analysis — not part of the static publishing pipeline.
| name | video-downloader |
| description | Download a video from YouTube, Instagram, TikTok, Twitter/X, Facebook, Vimeo (and 1000+ yt-dlp sites) to local disk, and optionally transcribe it. Standalone utility for repurposing and analysis — not part of the static publishing pipeline. |
| allowed-tools | Read, Bash |
| area | Productivity |
| use_for | Download a video from a URL (YouTube/IG/TikTok/X/FB/Vimeo/+) to local disk and optionally transcribe it — a standalone repurposing/analysis utility |
| deps | {"files":[],"env":[]} |
| Agent | Version | Last Changed |
|---|---|---|
| Link | v1.0.1 | May 28, 2026 |
Description: Download a video from a URL to local disk and optionally transcribe it — standalone utility.
v1.0.1 — May 28, 2026
scripts/download_video.py + $CLAUDE_PLUGIN_ROOT path. Cowork (not Claude Code) is the runtime and doesn't expose CLAUDE_PLUGIN_ROOT, and link-skills has no bundled-runtime-script convention. Reworked to run yt-dlp inline (python3 -m pip install yt-dlp → python3 -m yt_dlp) in the skill's Bash block — the same local-glue pattern content-generator uses for its inline Python. No plugin-relative paths, no shipped .py.v1.0.0 — May 28, 2026
meeting-analyzer. No brand context required.A standalone utility — no brand context, no Notion, no dashboard logging. Use it whenever a video URL needs to come down to local disk for repurposing (clip → static post), analysis (transcribe → summarize), or archiving. Everything runs inline in the Cowork bash sandbox — there is no bundled script and no gateway tool (yt-dlp is a local binary; the serverless gateway can't run it).
Do not use this for the static social pipeline (social-calendar → content-creation → creative-designer → social-publisher). It stands alone.
yt-dlp is a Python package — install it on demand and run it as a module (no PATH dependency, no bundled file). Substitute the real URL and (optionally) OUT:
OUT="${OUTPUT_DIR:-/tmp/downloads}"; mkdir -p "$OUT"
python3 -m pip install --quiet --upgrade yt-dlp 2>/dev/null || pip install --quiet --upgrade yt-dlp
# Best-quality MP4 (merges video+audio when ffmpeg is present)
python3 -m yt_dlp -f "bestvideo[ext=mp4]+bestaudio/best[ext=mp4]/best" --merge-output-format mp4 \
--no-playlist --no-warnings -o "$OUT/%(title).80s.%(ext)s" "VIDEO_URL"
Variants:
| Need | Tweak |
|---|---|
| Cap quality (e.g. 720p) | -f "bestvideo[height<=720][ext=mp4]+bestaudio/best[height<=720]" |
| Audio only (MP3) | -x --audio-format mp3 (drop the -f / --merge-output-format flags) |
| Output directory | set OUT (default /tmp/downloads) |
If the merge step fails (no ffmpeg in the sandbox), retry with a single-file format that needs no merge:
python3 -m yt_dlp -f "best[ext=mp4]/best" --no-playlist --no-warnings -o "$OUT/%(title).80s.%(ext)s" "VIDEO_URL"
Report the resulting file path (most recent file in $OUT):
ls -t "$OUT" | head -1
Supported: YouTube, Instagram, TikTok, Twitter/X, Facebook, Vimeo, and 1000+ other yt-dlp sites. If pip/network is unavailable in the sandbox, say so rather than failing silently.
Only when a transcript is wanted, and only if ffmpeg + whisper are available (both are heavy — guard for them):
command -v ffmpeg >/dev/null || { echo "ffmpeg not available here — cannot transcribe"; exit 0; }
ffmpeg -i "/path/to/video.mp4" -ar 16000 -ac 1 -y /tmp/video-audio.wav
python3 -m pip install --quiet --upgrade openai-whisper 2>/dev/null
python3 -c "import whisper; print(whisper.load_model('base').transcribe('/tmp/video-audio.wav')['text'])"
Save the transcript next to the video (<basename>.txt) or print it. A transcript can feed meeting-analyzer (action items) or content-creation (repurposing into posts).
rm -f /tmp/video-audio.wav when done. Leave the downloaded video unless the user asked for transcript-only.
CLAUDE_PLUGIN_ROOT). If pip / network / ffmpeg aren't available in the sandbox, report the limitation rather than failing silently.Write persona-targeted marketing copy — landing pages, emails, ad copy, blog posts, social media copy for any active brand
Analyze organic content performance for any active brand — your own published posts (engagement by topic, format, persona, angle, hook archetype, Direction) plus competitor content benchmarking — and produce a Performance Brief that feeds the social calendar.
Plan a weekly social media content calendar for any active brand — Static Mode (14 posts across LinkedIn, Facebook, Instagram) or YouTube-First Mode (one weekly video + platform clips), selected by brand.md Content Strategy. Runs weekly on Sunday cron schedule.
Daily live-trend scan for any active brand — surfaces timely topics, launches, news, and community pain points in the brand's niche, scored for relevance and timeliness, deduplicated against a rolling log, and written as candidate topics the social calendar can pull.
Daily automated content production — generate copy and images from Notion Social Calendar, publish to Zernio API, update Notion, notify Slack. Runs daily on cron schedule.
Visual design and asset creation — social media graphics, HTML/CSS mockups, image generation, text overlays and branding for any active brand