deepseek-harness
deepseek-ai
DeepSeek Harness: Everything is a Plugin.
PROJECT TOPICS
PROJECT README
DeepSeek Harness(DSH)生态的媒体下载插件。检测并下载 Bilibili / YouTube / 抖音(Douyin)/ 小红书(Xiaohongshu)等站点的视频与音频。
Media downloader for the DeepSeek Harness ecosystem. Detects and downloads video/audio from Bilibili, YouTube, Douyin and Xiaohongshu — directly from your agent conversation.
DSH 插件生态里有浏览器自动化插件,但没有媒体下载器工具。本插件填补了这个缺口:把 Chrome MV3 扩展(video-downloader-extension)里的媒体识别逻辑从浏览器沙箱中剥离出来,变成纯 Node 函数,并以三个 agent 可调用的工具暴露给模型。
The DSH plugin ecosystem has browser-automation plugins but no media-downloader tool. This plugin fills that gap: it lifts the media-detection logic out of the Chrome MV3 extension's browser sandbox into pure Node functions, exposed as three agent-callable tools.
video_analyze — 只做分析、不下载。返回 JSON:{ isMedia, site, quality, ext, mimeHint }。按 hostname 识别站点(bilibili / youtube / douyin / xiaohongshu / other)。
Pure analysis, no download — classify a URL's media-ness, site, quality, container and mime hint.
video_download — 用浏览器风格 User-Agent + Referer 请求媒体 URL,fs.createWriteStream 流式写入磁盘(默认 ./downloads/<sanitized-filename>),跟随重定向(最多 5 次),60 秒超时(并响应执行取消)。返回 { savedTo, bytes, quality, ext, elapsedMs }。
Stream a media URL to disk with redirect handling and a 60s timeout.
video_quality_parse — 纯正则质量/扩展名分析:{ quality, ext }。无网络请求。
Pure regex quality + extension extraction, zero network access.
零构建步骤——纯 ESM,发布包即运行时代码。 Zero build step — pure ESM, the published package is the runtime code.
纯函数全部导出,可直接用于测试与二次开发。
The pure core (isMedia, getQuality, getExt, detectSite) is exported for testing.
# 从 npm 安装 / from npm
dsh plugin --profile myprofile add dsh-video-downloader
# 从 GitHub 安装(锁定 commit 以保证供应链安全) / from GitHub (lock the commit)
dsh plugin --profile myprofile add github:yourname/dsh-video-downloader#<sha>
用自然语言让 agent 执行:
"分析这个页面里的媒体:https://www.bilibili.com/video/BV1xx411c7mD" "把这几个链接的 1080P 视频下载到 ./downloads 目录"
The agent calls the tools with:
{ "url": "https://bilivideo.com/xxx.m4s" } // video_analyze
{ "url": "https://bilivideo.com/xxx.m4s", "outPath": "./downloads/clip.mp4" } // video_download
{ "url": "https://example.com/video.mp4?height=720" } // video_quality_parse
video_download 也接受可选的 headers 对象合并到默认的浏览器请求头中。
video_download also accepts an optional headers object merged over the browser-like defaults.
| 工具 / Tool | 参数 / Parameters | 返回 / Returns |
|---|---|---|
video_analyze |
url (string, 必填) |
{ isMedia, site, quality, ext, mimeHint } |
video_download |
url (string, 必填), outPath (string, 可选), headers (object, 可选) |
{ savedTo, bytes, quality, ext, elapsedMs } |
video_quality_parse |
url (string, 必填) |
{ quality, ext } |
npm install
npm test # node:test,只测纯函数,不发网络请求 / pure functions only, no network
只下载你有权获取的内容。 本插件只负责媒体识别与文件流式保存,不绕过任何 DRM、登录墙或平台付费内容。请遵守各平台的服务条款与著作权法。
Only download content you have rights to. This plugin performs media detection and file streaming only — it does not bypass DRM, login walls or paid content. Respect each platform's terms of service and copyright law.
MIT
CLASSIFICATION EVIDENCE
系统优先读取 GitHub Topics,再与站内分类词典和词根规则比对。当前命中: 无有效分类标签。