video-subtitle-skill

为视频/音频自动生成字幕，支持多语言识别、翻译、说话人分离、字幕烧入视频

Security Scan

The skill appears to implement a legitimate video→subtitle workflow using SenseAudio, but the package metadata omits the required SENSEAUDIO_API_KEY and the runtime instructions include an echo of that secret (risking leakage); these mismatches and minor scope issues merit caution before installing.

Purpose & Capability

The skill's description and code use the SenseAudio ASR API and therefore require an API key and ffmpeg/ffprobe, but the registry metadata declares no required environment variables or required binaries. That omission is inconsistent: a subtitle generator that calls an external ASR certainly needs an API key and system media tools.

Instruction Scope

Runtime instructions tell the agent to run the included Python script (expected) and to read the generated text file for summarization (expected). However the SKILL.md first step prints the API key with echo (echo "SENSEAUDIO_API_KEY=$SENSEAUDIO_API_KEY") which can expose the secret in logs/outputs — this is an unnecessary secret leak. Instructions otherwise stay within the stated purpose and reference only the input file and generated outputs.

Install Mechanism

There is no install spec (instruction-only), which minimizes installer risk. However the skill ships an executable script (scripts/video_subtitle.py) but does not declare any installation steps for Python dependencies beyond documenting 'requests' and system deps in README/USAGE — this is a minor inconsistency (code is present but no install automation).

Credentials

The code requires a single external credential (SENSEAUDIO_API_KEY) and expects ffmpeg/ffprobe on PATH, but the registry metadata lists no required env vars or binaries. Requiring an API key is proportionate to the purpose, but failing to declare it in metadata is a security/operational gap. Additionally, the SKILL.md's practice of echoing the API key risks exposing it; this is disproportionate and unnecessary.

Persistence & Privilege

The skill does not request always:true, does not declare edits to other skill configs, and has no special persistence or elevated privileges. It runs as a one-off script and outputs local files.

What to consider before installing

This skill largely does what it claims (extract audio, call SenseAudio, write SRT/VTT/TXT, optionally burn subtitles), but there are important mismatches you should address before use: 1) The package metadata does not declare the required SENSEAUDIO_API_KEY or ffmpeg/ffprobe even though the script needs them — treat that as an omission, not a feature. 2) The SKILL.md echoes the API key to stdout which can leak the secret to logs or other observers; remove that echo or avoid running it. 3) Review the included scripts (scripts/video_subtitle.py) for any code paths that post data to endpoints you don't expect (the script appears to call only https://api.senseaudio.cn). 4) Run the skill in an isolated environment (or with an API key with limited quota) until you confirm behavior. 5) If you plan to install, ensure ffmpeg/ffprobe are present and supply the SENSEAUDIO_API_KEY securely (do not paste it into chat or run the provided echo). If you want higher confidence, ask the author to update metadata to declare the env var/binaries and to remove the secret-echoing line.

视频字幕生成器 (Video Subtitle Generator)

基于 SenseAudio ASR API，为视频或音频文件自动生成字幕。

核心功能

1. 语音识别 — 自动识别视频/音频中的语音内容，生成带时间戳的字幕
2. 多语言支持 — 支持中文、英文、日文、韩文等 20+ 种语言
3. 字幕翻译 — 识别后可自动翻译成目标语言
4. 说话人分离 — 多人对话场景自动区分不同说话人
5. 字幕烧入 — 将生成的字幕直接烧入视频输出新文件
6. 多格式输出 — 支持 SRT / VTT / TXT / JSON 格式

使用方式

用户说出类似以下请求时触发此 Skill：

• "帮我给这个视频加字幕"
• "识别这个音频的内容并生成字幕"
• "把这个英文视频翻译成中文字幕"
• "帮我总结一下这个视频讲了什么"

执行步骤

第一步：检查 API 密钥

echo "SENSEAUDIO_API_KEY=$SENSEAUDIO_API_KEY"

如果 SENSEAUDIO_API_KEY 为空，必须先向用户询问，说明在 https://senseaudio.cn 注册获取。不要直接运行脚本让它报错。

第二步：运行脚本生成字幕

# 基础用法
python scripts/video_subtitle.py "/path/to/video.mp4" --output outputs/

# 指定语言 + 翻译
python scripts/video_subtitle.py "/path/to/video.mp4" --language zh --translate en

# 生成字幕并烧入视频
python scripts/video_subtitle.py "/path/to/video.mp4" --burn --font-size 28

# 说话人分离
python scripts/video_subtitle.py "/path/to/meeting.mp4" --model pro --speaker

注意：如果环境变量 SENSEAUDIO_API_KEY 已设置，无需 --senseaudio-api-key。

第三步：内容总结（如用户需要）

脚本会输出纯文本转写结果（*.txt）。如果用户需要视频内容总结，你（Claude）直接读取这个 txt 文件并总结，不需要调用外部 LLM。

第四步：返回结果

将生成的字幕文件路径（和烧入后的视频路径）返回给用户。

ASR 模型选择

模型	参数值	特点	适用场景
极速版	lite	毫秒级响应、30+ 语言	低成本批量转写
标准版	standard	功能全面、性价比高	通用转写、视频字幕（推荐）
专业版	pro	高精度、说话人分离	会议记录、访谈、多人场景
深度版	deepthink	智能纠错、方言增强	方言/术语较多的场景

环境要求

• Python 3.10+，依赖：requests
• 系统依赖：ffmpeg（音频提取和字幕烧入）、fonts-noto-cjk（中日韩字幕烧入需要）
• SENSEAUDIO_API_KEY — SenseAudio API 密钥（唯一需要的密钥）

参数说明

参数	说明	默认值
input	输入视频/音频文件路径（必填）	-
--output	输出目录	输入文件同级 subtitle_output/
--model	ASR 模型 (lite/standard/pro/deepthink)	standard
--language	音频语言代码 (zh/en/ja 等)	自动检测
--translate	翻译目标语言代码	不翻译
--speaker	启用说话人分离	否
--sentiment	启用情感分析	否
--burn	将字幕烧入视频	否
--font-size	烧入字幕字体大小	24
--format	字幕格式 (srt/vtt/both)	srt
--senseaudio-api-key	SenseAudio API 密钥	环境变量

输出文件

文件	说明
文件名.srt	SRT 格式字幕
文件名.vtt	VTT 格式字幕（需指定 --format vtt/both）
文件名.txt	纯文本转写结果
文件名_detail.json	详细识别结果（含时间戳、说话人等）
文件名_subtitled.mp4	烧入字幕的视频（需指定 --burn）

支持的语言代码

zh(中文) en(英文) ja(日文) ko(韩文) fr(法语) de(德语) es(西班牙语) pt(葡萄牙语) ru(俄语) it(意大利语) ar(阿拉伯语) yue(粤语) nl(荷兰语) id(印尼语) ms(马来语) th(泰语) tr(土耳其语) vi(越南语) 等