← 返回 Skills 市场
zhaoyta

video-remix

作者 zhaoyta · GitHub ↗ · v1.0.7 · MIT-0
cross-platform ⚠ suspicious
273
总下载
1
收藏
0
当前安装
8
版本数
在 OpenClaw 中安装
/install video-remix
功能描述
YouTube 视频处理助手 - Gemini 智能分析 + 下载 + 剪辑 + 配音 + 硬字幕合成 + 局域网 HTTP 分享
安全使用建议
This package appears to implement the advertised video download→transcribe→analyze→TTS→merge flow, but there are privacy-exposing details you should consider before installing or running it: - Browser profile risk: SKILL.md tells the agent to start a browser with profile "openclaw" and automate Gemini. That will use whatever cookies, sessions, and credentials are present in that profile (including any logged-in Google account). If you don't want your Google session or other profile data used, do not run with an existing profile — run with a fresh/ephemeral profile or in an isolated VM/container. - Autonomy and triggers: The skill is written to auto-trigger on specific keywords. If you enable auto-invocation, the agent may run the Gemini automation and downloads without an explicit per-run approval. Prefer explicit user invocation unless you trust it. - LAN sharing: The skill promises HTTP sharing on the LAN. Inspect the scripts that start servers to ensure they bind only to localhost or are protected; otherwise files could be exposed to your network. - LLM credentials and data exfiltration: The skill will send the YouTube URL and generated prompts to Gemini (via browser) and may also call OpenAI/Anthropic/local endpoints if configured. Be aware that transcripts and prompts (including video content/context) will be transmitted to those services. If you must keep content private, do not use remote LLMs or run the tool offline. - Running commands: The code invokes system binaries (yt-dlp, ffmpeg, whisper/faster-whisper, edge-tts). Review the download and encoding commands and ensure you install trusted builds (especially ffmpeg with libass) from your OS package manager or trusted taps. - Safety checklist before running: (1) Review the full scripts (video_remix.py, gemini_first_remix.py) to see exact network endpoints and any HTTP servers started; (2) Run inside an isolated environment (container/VM) or with a fresh browser profile; (3) Do not supply any API keys unless you understand how/where they're used and stored; (4) Confirm legal/compliance implications of downloading YouTube content. If you want, I can scan the remaining truncated files (scripts) for any code that starts a network listener, posts data to external endpoints, or persists secrets, and produce a tighter list of lines to inspect.
功能分析
Type: OpenClaw Skill Name: video-remix Version: 1.0.7 The skill bundle implements a comprehensive video processing pipeline including YouTube downloading, browser-based AI analysis, and local file sharing. It is classified as suspicious due to several high-risk capabilities: it instructs the agent to perform system-level package installations (e.g., `sudo apt install` in `SKILL.md`), utilizes browser automation to scrape content from Gemini, and launches a local HTTP server (`scripts/gemini_first_remix.py`) to expose the workspace to the local network. While these behaviors are aligned with the stated goal of automated video editing and sharing, the combination of shell execution, remote content fetching, and network service hosting creates a significant attack surface.
能力评估
Purpose & Capability
Files (downloader, transcriber, analyzer, tts, merger, script_gen) align with a YouTube remix workflow. The skill supports multiple LLM backends (Gemini via browser automation, OpenAI, Anthropic, local host) though no credentials are declared — this is plausible but the presence of LLM client code (openai/anthropic/requests) is broader than the SKILL.md's focus on Gemini-only automation.
Instruction Scope
SKILL.md requires immediate browser automation using browser(action="start", profile="openclaw") and then navigates to gemini.google.com, types prompts, reads responses and saves them to temp/gemini_result.json. Starting a browser with a named profile can expose cookies, auth tokens, and other sensitive data from the user's profile; the document does not justify or limit use of that profile. The instructions also mandate automatic triggering on keywords, which could cause the agent to run browser actions without explicit user consent each time those keywords appear.
Install Mechanism
No install spec in registry; SKILL.md provides explicit platform checks and pip/apt/brew install commands for dependencies (ffmpeg, yt-dlp, edge-tts). Install steps use standard package managers and package names; no arbitrary external archive downloads are requested.
Credentials
The skill declares no required env vars, but code reads standard proxy env vars (HTTP_PROXY/HTTPS_PROXY) and supports API keys for OpenAI/Anthropic/local LLMs (api_key/base_url passed in ScriptGenerator). The implicit reliance on an existing browser profile and optional LLM API keys are sensitive and not declared as required — this is reasonable design-wise but worth caution because credentials/session data could be accessed during browser automation.
Persistence & Privilege
always is false and the skill is user-invocable; autonomous invocation on keyword triggers is specified in SKILL.md (allowed by platform defaults). The real concern is operational: browser automation with a named profile and a promised "局域网 HTTP 分享" (LAN HTTP sharing) can expose_local files or services on the network. The skill does not request system-wide persistent privileges, but its behavior could nonetheless publish content to the LAN if scripts start a server.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install video-remix
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /video-remix 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.7
- No user-facing changes; documentation and implementation remain the same. - Version update with no detected file changes.
v1.0.6
### video-remix 1.0.6 更新日志 - 大幅升级 Gemini 分析模板:输出 JSON 增加抖音爆款标题(douyin_title)和标签(douyin_tags),文本提示支持“期望总时长”参数,引导更优质、新媒体风格输出。 - 明确分阶段流程和依赖检查(阶段 0):强制本地依赖自检(ffmpeg, yt-dlp, edge-tts, libass)并给出适配安装命令。 - 输出说明补全:详细列出所有产物(mp4、mp3、srt、HTTP 地址)及存储目录。 - 片段选取标准更严格,新增避坑项(不要选带字幕片段)。 - 文档结构优化,内容流程更清晰,便于理解和实践。
v1.0.5
**Video Remix Skill 1.0.5 Changelog** - Major update: Now adopts a 2-stage process with explicit phase separation for improved reliability. - Added detailed workflow for fully automated Gemini analysis using browser tools, including code snippets for browser interaction. - Clearly documents pre-checks and error handling between Gemini JSON generation and video processing. - Updated and expanded usage instructions, emphasizing the requirement to first obtain and validate `gemini_result.json` before running the main processing script. - Refined and clarified all workflow steps, making integration and troubleshooting easier for all users.
v1.0.4
**Changelog for video-remix v1.0.4** - 增加 Gemini 提示词要求,明确:不得选择出现在片段中的人脸,以及不得选择带有明显水印/Logo 的片段。 - 强调每次必须完整复制“Gemini 提示词模板”全部内容,禁止删减或简化,防止输出质量下降。 - FFmpeg 硬字幕检查命令细化,增加 2>&1 保证全部输出被检查。 - 注意事项与高风险点更新,更醒目提示严格流程和不可删减信息。 - 其他文档内容小幅补充优化,增强流程准确性和易用性。
v1.0.3
- Major update: the Gemini prompt is now integrated and simplified directly within the documentation for easier use. - Added a clear, copy-paste Gemini prompt template, including detailed instructions and JSON output requirements. - Provided real-world Gemini input/output examples for different types of YouTube videos. - Refined the workflow section for better clarity; renamed to V5 and described correct browser steps. - Added a quick FFmpeg+libass compatibility check and updated installation guidance. - Removed redundant or separate prompt files (such as GEMINI_PROMPT_FULL.md).
v1.0.2
- Removed 16 files, including all core processing scripts and documentation (README.md, add_subs.py, gemini_edit.py, etc.). - Significantly reduced the codebase, leaving only the main skill description file. - The skill's documentation and implementation scripts are no longer included in this version.
v1.0.1
Initial release with full Gemini-first YouTube video remix workflow. - Introduced Gemini智能分析 workflow: 自动分析视频精彩片段、生成配音文案,输出时间戳和文案JSON。 - 支持 YouTube 视频下载、剪辑、配音(Edge-TTS)、字幕生成与 FFmpeg 硬字幕合成,一站式处理。 - 自动检测并对齐文案字数与视频片段时长,过长自动加速配音对齐视频,字幕适配视频。 - 局域网 HTTP 服务自动分享所有输出文件和硬字幕视频,便于快速访问和分发。 - 全面升级文档,详细说明流程、触发词、命令行用法、可用 TTS 声音与系统依赖。
v1.0.0
Initial public release. - Introduces the video-remix skill with detailed guidance on skill structuring and documentation practices. - Provides clear templates for organizing workflows, tasks, references, and resources. - Includes placeholders for descriptive content, intended usage scenarios, and example resource directories. - Sets up a foundation for future expansion with implementation guidelines and modular structure suggestions.
元数据
Slug video-remix
版本 1.0.7
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 8
常见问题

video-remix 是什么?

YouTube 视频处理助手 - Gemini 智能分析 + 下载 + 剪辑 + 配音 + 硬字幕合成 + 局域网 HTTP 分享. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 273 次。

如何安装 video-remix?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install video-remix」即可一键安装,无需额外配置。

video-remix 是免费的吗?

是的,video-remix 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

video-remix 支持哪些平台?

video-remix 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 video-remix?

由 zhaoyta(@zhaoyta)开发并维护,当前版本 v1.0.7。

💬 留言讨论