菜单
Consolidated single-source voiced-video workflows for speech condensing, talking-head subtitles, and end-to-end caption/edit orchestration. Use when video-edit-agent routes a single voiced video to mode=speech-condense, mode=talking-head-subtitles, or mode=video-pipeline.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/wwwzhouhui/skills_collection --skill video-speech-workflows命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Checking and provisioning the machine's environment for the video-agent-kit plugin — probing for ffmpeg/ffprobe that actually carry the encoders and filters we render with (libx264/aac/libmp3lame, libass, libfreetype), python packages (mcp, cv2, numpy, PIL, jieba, fontTools, scenedetect) in the interpreter Claude Code really uses, a CJK font whose cmap truly covers Chinese, and which speech channel is in play (under ZCode the official channel needs no local key) — then installing what is missing (China mirrors by default) or telling the user exactly what to install. Works on macOS, Windows and Linux. Use BEFORE the first edit run on a new machine, when the session-start hook reports 缺失依赖, or whenever a tool fails in an environment-shaped way — ModuleNotFoundError, `ffmpeg not found`, `Unknown encoder 'libx264'`, `No such filter: 'subtitles'`, `no font file found for family`, 字幕烧成豆腐块/方块, speech_transcribe 超时或 401. Also use when asked to 检查环境 / 装依赖 / 配国内源 / setup / doctor.
Global controller for video editing tasks. Use as the default entrypoint for any video edit request; it discovers materials, classifies the task, routes to specialized video-edit skills when available, and enforces the common file contract, timeline validation, preview rendering, QC, repair loop, and final report.
Specialized workflow for multi-material video assembly. Use after video-edit-agent classifies a task as combining, selecting, ordering, de-duplicating, or shaping multiple user-provided videos/audio/images into one coherent finished video, montage, recap, social-media cut, or story.
正在显示 SKILL.md
| name | video-speech-workflows |
| description | Consolidated single-source voiced-video workflows for speech condensing, talking-head subtitles, and end-to-end caption/edit orchestration. Use when video-edit-agent routes a single voiced video to mode=speech-condense, mode=talking-head-subtitles, or mode=video-pipeline. |
This skill contains the specialized single-source speech/video workflows that sit beside video-edit-assembly.
Choose exactly one mode from the caller's args, then read and follow the matching workflow document:
| Mode | Workflow |
|---|---|
speech-condense | workflows/speech-condense.md |
talking-head-subtitles | workflows/talking-head-subtitles.md |
video-pipeline | workflows/video-pipeline.md |
If the args do not name a mode, infer the narrowest matching one:
speech-condense for one long unscripted recording that must be shortened, tightened, or cut to a target length.talking-head-subtitles for straight captioning, subtitle repair, speaker-labelled captions, karaoke highlighting, bilingual lines, or subtitle placement/legibility work on one rendered video.video-pipeline when a single voiced video needs an end-to-end stage plan that may combine condensing/editing with captions, speaker differentiation, karaoke, translation, or terminology notes.video-edit-assembly or the common video-edit-agent workflow first. Subtitle the rendered output afterward so cue times match the final video.video_watch_segment(video_path=...) can directly inspect tight windows without a prior full video_ingest; reserve full ingest for workflows that genuinely need dense whole-video contact sheets.