| name | video-summarizer |
| description | Use when the user provides a video or audio URL and asks to summarize, extract subtitles, create video notes, 视频总结, 视频笔记, or summarize content from Bilibili, YouTube, Douyin, Xiaohongshu, TikTok, or yt-dlp supported sites. |
Video Summarizer
Core Rule
Extract the transcript first, then summarize from evidence. Do not invent content when extraction fails.
This skill is a Codex-native port of keepongo/video-summarizer at upstream commit 481a772.
Supported Inputs
- Bilibili:
bilibili.com/video/, b23.tv/, BV...
- YouTube:
youtube.com/watch, youtu.be/, youtube.com/shorts/
- Douyin:
douyin.com, v.douyin.com
- Xiaohongshu:
xiaohongshu.com, xhslink.com
- TikTok and many other sites through
yt-dlp
Workflow
- Choose dependencies from
references/dependencies.md.
- Run
scripts/video_subtitle.py with uv run python ...; add uv --group ... dependency groups only when the platform needs them.
- Parse the JSON output. If it contains
error, report the failure and the next concrete fix.
- Summarize only from
subtitle_text. If frames are present, embed selected images with absolute paths.
- Use the output format in
references/output-format.md.
Commands
Bilibili or direct-platform extraction:
uv run python .agents/skills/video-summarizer/scripts/video_subtitle.py "<VIDEO_URL>"
YouTube:
uv run --group video-youtube python .agents/skills/video-summarizer/scripts/video_subtitle.py "<VIDEO_URL>"
TikTok, generic sites, or yt-dlp fallback:
uv run --group video-ytdlp python .agents/skills/video-summarizer/scripts/video_subtitle.py "<VIDEO_URL>"
OpenAI Whisper API fallback:
OPENAI_API_KEY=... uv run --group video-ytdlp --group video-whisper-api python .agents/skills/video-summarizer/scripts/video_subtitle.py "<VIDEO_URL>"
Local Whisper fallback is heavy. Use it only when the user explicitly wants audio transcription and accepts model downloads:
uv run --group video-ytdlp --group video-whisper-local python .agents/skills/video-summarizer/scripts/video_subtitle.py "<VIDEO_URL>"
Clear cached transcript and screenshots:
uv run python .agents/skills/video-summarizer/scripts/video_subtitle.py --clear-cache
Configuration
- Copy
references/config.example.json to .agents/skills/video-summarizer/config.json for local settings.
- Use
references/config.whisper-api.example.json or references/config.whisper-local.example.json when enabling Whisper.
- Read
references/setup.md when initializing this skill on a new machine or checking environment readiness.
- Keep
config.json, cookies, cache, and screenshots untracked.
- Prefer
OPENAI_API_KEY for OpenAI Whisper API instead of writing a key into config.
Failure Handling
- Missing dependency: rerun with the matching
uv run --group ... command.
- No subtitles: ask whether to enable Whisper fallback.
- Frame extraction missing: continue with text-only output unless the user specifically needs images.
- Platform/cookie block: explain the platform limitation and point to
references/dependencies.md.
