video-transcribelisted

# Video Transcribe Use this skill when the user asks to understand, transcribe, summarize, or visually analyze a video/audio source. ## Mode Selection - **Full note / no omission**: user says 全量、完整、不要看视频、替代看视频、图文笔记, or asks about completeness. Download the source, transcribe the whole audio, extract frames, and produce a coverage-checked note. - **Audio transcript**: user explicitly asks for 字幕、转录、他说了什么、transcribe. Produce a timestamped transcript or transcript-backed summary. - **Visual analysis**: user explicitly asks about 画面、截图、视觉、展示了什么. Extract frames and analyze visuals; skip Groq unless speech content matters. - **Quick summary**: user only wants a brief answer. Summarize, but still state whether the result is transcript-backed, frame-backed, or both. When intent is unclear, default to transcript + keyframes for short videos and ask before spending API quota on long videos. ## Core Rules - For long-form technical content, prefer Groq `whisper-large-v3` with `response_format=verbose_json`; do not use plain `text` when timestamps or completeness checks matter. - Do not force `language=zh` for non-Chinese videos. Use the detected language or set `--language en` for English technical videos. - Use a terminology prompt for product/tool names such as `Codex`, `Remotion`, `Supabase`, `Typefully`, `TestFlight`, `Vercel`, `Claude Code`. - Source timestamps are an outline, not proof. Verify every listed timestamp has transcript coverage before claiming the note can replace