调用senseaudio asr的课堂转译助手，将英文课堂录音、讲座视频、组会音频等内容自动转写为文本，生成中文总结

接收课堂录音、讲座音频或视频文件（视频会先抽取音轨），调用 SenseAudio HTTP ASR API 进行英文转录，可选直出中文翻译；随后整理为结构化 Markdown 学习笔记，包含摘要、关键概念、术语表、时间轴与复习问题，生成到桌面，并支持导出到 Notion 或保存到 Obsidian vault。

Security Scan

The skill is internally consistent with its stated purpose: it uploads provided or page-extracted audio to SenseAudio for ASR, optionally uses OpenAI/Notion, and writes Markdown/Obsidian outputs — it requests only the SenseAudio API key as primary credential.

Purpose & Capability

Name/description match what the files do: script accepts audio/video or a page URL, extracts audio (ffmpeg for video), calls the SenseAudio HTTP ASR endpoint, optionally calls OpenAI-compatible summary provider and/or Notion API, and writes Markdown/Obsidian files. Required binary (python3) and declared env SENSEAUDIO_API_KEY are appropriate for the stated functionality.

Instruction Scope

SKILL.md and script instruct the agent to fetch page HTML (when --page-url is used), download media, extract audio, upload the audio to SenseAudio, and optionally push content to Notion or write into an Obsidian vault. These actions are within the stated scope but have privacy implications: user audio (and any media downloaded from a page) will be transmitted to external services (SenseAudio and optionally Notion/OpenAI). The instructions do not appear to read unrelated local config or credentials.

Install Mechanism

No install script that pulls arbitrary binaries; dependencies are minimal (requests). This is an instruction-only skill with a Python script; risk from install mechanism is low. Users must manually install dependencies and ffmpeg for video handling per docs.

Credentials

Primary required env var is SENSEAUDIO_API_KEY, which is justified. OPENAI_API_KEY, OPENAI_BASE_URL, and NOTION_TOKEN are optional and only needed for explicit optional functionality. No unrelated credentials are requested.

Persistence & Privilege

always is false and the skill does not request permanent platform presence or attempt to modify other skills. It writes output files to user-specified directories and optionally to a supplied Obsidian vault path, which is expected behavior.

Assessment

This skill behaves as advertised but you should be aware of privacy and data-flow consequences before installing: any audio you provide (or media downloaded from a web page via --page-url) will be uploaded to SenseAudio for transcription, and summaries may be sent to an OpenAI-compatible endpoint if you enable that provider; exporting to Notion requires giving a Notion token which grants the skill access to create pages. If your recordings contain sensitive or confidential information, do not run them through this skill or review SenseAudio/Notion/OpenAI privacy and retention policies first. Also ensure ffmpeg is installed locally for video inputs and that the output directory / Obsidian vault paths are correct and writable. If you want higher assurance, review the full script locally (scripts/classroom_note_translator.py) before running — the code is small and readable and the scanned package shows no pre-scan injection signals.

课堂听译笔记官

当用户希望把英文课堂录音、讲座、组会、YouTube 学习音频/视频整理成中文笔记时，使用这个 skill。 触发（更明确）：

• 用户明确提到“总结”课堂内容
• 用户明确提到“转译/翻译”课堂音视频内容
• 用户明确提到“做中英对照笔记”或“整理课堂笔记”
• 用户提供网页链接并希望从页面里的音视频提取后再做总结/转译 非触发：
• 仅做逐字转写、且不需要总结和中英对照时 适用场景：
• 国外课堂听讲太快，想保留英文原文并生成中文笔记
• 学术报告、课程录音、线上讲座需要整理成复习材料
• 希望得到 Markdown 文档，并进一步导入 Notion 或 Obsidian

你应该做什么

1. 收集或确认以下输入：
   • 输入来源（三选一）：音频路径 / 视频路径 / 页面链接（从页面提取媒体）
   • 课程标题或主题（可选）
   • 是否需要说话人分离
   • 是否需要句级或字级时间戳
   • 输出目录（必填，使用用户指定目录）
   • 是否导出到 Notion
   • 是否写入 Obsidian vault
2. 运行脚本：

python3 "{baseDir}/scripts/classroom_note_translator.py" \
  --audio "<AUDIO_PATH>" \
  # 或 --video "<VIDEO_PATH>" \
  # 或 --page-url "<PAGE_URL>" \
  --title "<TITLE_OR_TOPIC>" \
  --model sense-asr-pro \
  --language en \
  --target-language zh \
  --timestamps segment \
  --speaker-diarization \
  --summary-provider local \
  --output-dir "<USER_SPECIFIED_OUTPUT_DIR>"

3. 如果用户想导出到 Notion，则追加：

  --export-notion \
  --notion-parent-page-id "<NOTION_PAGE_ID>"

4. 如果用户想写入 Obsidian，则追加：

  --export-obsidian \
  --obsidian-vault "/absolute/path/to/your/vault" \
  --obsidian-folder "Lecture Notes"

输出内容

脚本每次都会同时生成：

• *_transcript.json：SenseAudio 原始响应
• *_summary.md：结构化中文学习总结（Markdown）
• *_bilingual.txt：英文原文 + 中文翻译拼接文本

Markdown 文档默认包含：

• 标题与元信息
• 中文摘要
• 关键要点
• 术语表
• 带时间轴的内容梳理
• 复习问题
• 英文原文摘录

参数选择建议

• 默认优先用 sense-asr-pro：适合课堂、会议、复杂语音环境。
• 若追求更低成本，可改为 sense-asr。
• 若只要快速粗转写，不需要翻译/时间戳/说话人分离，不建议用本 skill，改用轻量转写更合适。
• 对于长音频，先切片再分别处理；SenseAudio 官方文档建议单文件不超过 10MB。

环境变量

• SENSEAUDIO_API_KEY：必填，SenseAudio API Key
• OPENAI_API_KEY：可选，仅当 --summary-provider openai-compatible 时需要
• OPENAI_BASE_URL：可选，OpenAI 兼容接口地址
• NOTION_TOKEN：可选，导出到 Notion 时需要

本地依赖

• 若使用 --video，需本机安装 ffmpeg（用于抽取音轨）

注意事项

• SenseAudio 文档显示，sense-asr 与 sense-asr-pro 支持翻译、句级/字级时间戳、说话人分离与情感分析；HTTP 接口为 POST https://api.senseaudio.cn/v1/audio/transcriptions，采用 multipart/form-data 与 Bearer Token。
• 若使用 --page-url，脚本会先抓取页面 HTML，优先提取 og:video/og:audio 和 <video>/<audio>/<source> 的 src，下载后再处理；部分站点（需登录、DRM、动态脚本拼接媒体地址）可能提取失败。
• 若 Obsidian vault 不可写，则只生成 Markdown 文件，由用户自行导入。
• 若未配置 Notion token，则跳过 Notion 导出。
• 若用户没有给出标题，可由文件名自动生成。