Name: video-parse-transcribe
Author: YT-smart

Imported from YT-smart/video-parse-transcribe (SKILL.md). Install upstream with npx skills add YT-smart/video-parse-transcribe. Copyright stays with the author.

Video Parse Transcribe

Core Workflow

Use the bundled script with paths relative to this skill folder. Do not hard-code the creator's local absolute path.

From the skill directory:

python scripts/video_parse_transcribe.py --url "<share text or URL>" --mode parse

python scripts/video_parse_transcribe.py --url "<share text or URL>" --mode both --model tiny

If running from another directory, resolve the script path from the current skill folder instead of using a machine-specific path.

Modes:

parse: return JSON metadata only, including video_url, cover_url, title, author, and image/live-photo data when available.
both: parse the share link, download the video to a temporary directory, extract audio with ffmpeg, transcribe with local Whisper, and print raw transcript text.
transcribe: use when the input URL is already a direct media/play URL.

Default behavior:

Whisper model is tiny for speed.
Temporary video/audio files are automatically cleaned up.
No workspace output folder is created unless --keep-temp is passed.
Use --json only when both parse metadata and transcript are needed in one payload.

Response Rules

For transcript/copy requests:

Run --mode both --model tiny.
Treat script output as raw ASR, not final copy.
Correct obvious ASR errors using context before replying.
- Common examples: 等于 -> 抖音, 克服 -> 客服, 利益变 -> 裂变, 富盘 -> 复盘.
- Keep the speaker's wording and structure. Do not rewrite into a polished article unless the user asks.
Output only the corrected文案 when the user asks for文案/逐字稿. Do not include file paths, tool logs, or process notes unless requested.

For parse-only requests:

Run --mode parse.
Return the useful fields: title, author, video URL, cover URL, image URLs if present.
Mention that platform media URLs are usually time-limited.

Platform Policy

Advertise and rely on stable/common platforms only:

Douyin / 抖音
Xiaohongshu / 小红书
Bilibili / B站
Kuaishou / 快手
Direct media or play URLs, such as .mp4, .m3u8, or already-resolved platform play URLs

Do not advertise broad support for every parser inherited from parse-video-py. Other platforms may be tried as best effort only when the user explicitly asks, but report instability clearly.

Known boundaries:

Douyin app short links like https://v.douyin.com/... work.
Douyin PC links like https://www.douyin.com/jingxuan?modal_id=... work in current testing.
Already-resolved play URLs such as https://www.douyin.com/aweme/v1/play/... should be treated as direct media URLs instead of parsed as share pages.
Image-only posts may return no video_url; report images instead of attempting ASR.

Dependencies

The script vendors parse-video-py under scripts/vendor/parse_video_py, so it does not need the external parse_video MCP tool.

Runtime requirements:

Python 3.10+
ffmpeg on PATH for transcription
Python packages used by the vendored parser: httpx, fake-useragent, pyyaml, parsel, lxml, jmespath, aiohttp
Python package for transcription: whisper

If imports fail, install the missing package in the active Python environment. If network access fails in the sandbox, rerun the same command with escalation.

Install video-parse-transcribe and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

{ "ocm": "1", "id": "yt-smart-video-parse-transcribe-video-parse-transcribe", "kind": "skill", "name": "video-parse-transcribe", "description": "Parse video share links and extract spoken copy from short videos. Use for stable common platforms only: Douyin, Xiaohongshu, Bilibili, Kuaishou, and direct media or play URLs. Use for video parsing, metadata extraction, no-watermark links, transcription, subtitles, word-for-word copy, or a combined parse-and-transcribe workflow. Uses a vendored parse-video-py parser and local Whisper tiny, then Codex corrects obvious ASR typos before replying.", "publisher": "YT-smart", "version": "1.0.0", "capabilities": { "domains": [ "general" ], "tags": [ "skill-md", "github" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Parse video share links and extract spoken copy from short videos. Use for stable common platforms only: Douyin, Xiaohongshu, Bilibili, Kuaishou, and direct media or play URLs. Use for video parsing, metadata extraction, no-watermark links, transcription, subtitles, word-for-word copy, or a combined parse-and-transcribe workflow. Uses a vendored parse-video-py parser and local Whisper tiny, then Codex corrects obvious ASR typos before replying." ], "primary": false, "metadata": { "source": { "provider": "github", "repository": "https://github.com/YT-smart/video-parse-transcribe", "path": "SKILL.md", "ref": "35085f088eed89365ab68cb8cfe01aea02f525ce", "url": "https://github.com/YT-smart/video-parse-transcribe/blob/35085f088eed89365ab68cb8cfe01aea02f525ce/SKILL.md", "key": "YT-smart/video-parse-transcribe/SKILL.md" } }, "instructions": "# Video Parse Transcribe\n\n## Core Workflow\n\nUse the bundled script with paths relative to this skill folder. Do not hard-code the creator's local absolute path.\n\nFrom the skill directory:\n\n```powershell\npython scripts/video_parse_transcribe.py --url \"<share text or URL>\" --mode parse\n```\n\n```powershell\npython scripts/video_parse_transcribe.py --url \"<share text or URL>\" --mode both --model tiny\n```\n\nIf running from another directory, resolve the script path from the current skill folder instead of using a machine-specific path.\n\nModes:\n\n- `parse`: return JSON metadata only, including `video_ur", "cost": { "context_tokens": 829 } }

video-parse-transcribe

About