Ai Meeting Helper

会议纪要生成器 - 自动将会议录音转为结构化纪要

Security Scan

The skill's code matches its stated purpose (audio→transcript→LLM summary) and only needs an OpenAI API key, but there are metadata inconsistencies and some install choices that warrant caution.

Purpose & Capability

The code, SKILL.md, and skill.json all implement a meeting-transcription-and-summary tool that uses OpenAI Whisper and an LLM — requesting an OPENAI_API_KEY is coherent with that purpose. However, the top-level registry metadata provided to you (which claimed no required env vars / no primary credential) contradicts the included skill.json and SKILL.md that require OPENAI_API_KEY. The mismatch is unexpected and should be resolved before trusting the package metadata.

Instruction Scope

Runtime instructions and the Python code stay within the declared purpose: they read local audio files, call OpenAI APIs for transcription and summarization, format results, optionally backup original audio to a local .ai_meeting_backup/ directory, and write outputs. The instructions do not reference unrelated system paths, other credentials, or external endpoints beyond the OpenAI client.

Install Mechanism

Installation is a local install.sh that runs pip3 install openai python-dotenv (no virtualenv) and creates a .env template and backup/log directories in the skill folder. This uses a public PyPI package (openai) — not high risk by itself — but the script installs packages system-wide (no venv) which may be undesirable. No downloads from arbitrary URLs or obfuscated installers were found.

Credentials

The code requires a single credential (OPENAI_API_KEY), which is proportionate to the stated functionality. However, the provided registry metadata (at the top of this report) claims no required env vars/primary credential while skill.json and SKILL.md declare OPENAI_API_KEY required — that inconsistency is suspicious and could indicate sloppy packaging or metadata manipulation.

Persistence & Privilege

The skill does not request 'always: true' or elevated agent privileges. It creates and writes files only within its own skill directory (.env, .ai_meeting_backup/, .ai_meeting_logs/) and does not modify other skills or global agent config. Uninstall.sh offers to remove these files.

What to consider before installing

What to consider before installing: - Metadata mismatch: the skill.json and SKILL.md require OPENAI_API_KEY, but the top-level registry metadata claimed no required env vars. Treat that inconsistency as a red flag — confirm which metadata is authoritative before trusting automated installers. - Review code yourself (you have the files): the Python script uses only the OpenAI client and local file I/O; no hard-coded external endpoints or obfuscated code were found. If you don't review, install only in an isolated environment. - Install in a virtual environment or sandbox rather than system-wide (the install.sh runs pip3 install without a venv). This avoids contaminating global Python packages. - Backups: the tool will copy original audio into .ai_meeting_backup/. Those files can contain sensitive conversations — ensure you control that directory, or disable backup if not needed, and securely delete backups when appropriate. - API key scope and monitoring: provide an OpenAI key with appropriate usage controls; monitor usage and billing. If possible use an account with limited permissions or monitoring alerts. - Trust and provenance: owner is unknown and there's no homepage. If you need high assurance, request more provenance or prefer a skill from a known source. If you decide to proceed: run install.sh only after inspecting it, execute the tool inside a dedicated virtualenv or container, and consider auditing network traffic or logs during first runs. If you need help verifying specific lines of code or testing in a sandbox, provide the environment and I can help step through it.

AI Meeting Helper - 会议纪要生成器

自动将会议录音转换为结构化会议纪要，包括行动项、决策点和待办事项。

功能

• 🎙️ 语音转文字：使用 OpenAI Whisper API 将会议录音转为文本
• 📋 自动总结：使用 LLM 分析对话内容，生成结构化纪要
• ✅ 提取要点：自动识别行动项、决策点、待办事项
• 📤 多格式输出：支持 Markdown、纯文本、JSON 格式
• 🔄 批量处理：一次处理多个会议录音文件
• 👀 预览模式：不实际生成文件，只显示预览
• ↩️ 撤销功能：自动备份原始文件，可撤销操作

快速开始

1. 安装技能

clawhub install ai-meeting-helper
cd ~/.openclaw/workspace/skills/ai-meeting-helper
./install.sh

2. 配置 OpenAI API

export OPENAI_API_KEY="your-api-key"

3. 使用示例

# 单个文件处理
ai-meeting-helper process meeting_recording.mp3 --output meeting_notes.md

# 批量处理文件夹
ai-meeting-helper batch ./meetings/ --output ./notes/ --format markdown

# 预览模式（不生成文件）
ai-meeting-helper process meeting.mp3 --preview

# 启用撤销功能
ai-meeting-helper process meeting.mp3 --output notes.md --backup

参数

参数	类型	必需	描述
input	路径	是	输入音频文件或目录
--output	路径	否	输出纪要文件路径（默认：当前目录）
--format	选项	否	输出格式：markdown/text/json（默认：markdown）
--model	字符串	否	Whisper 模型：tiny/base/small/medium/large（默认：base）
--llm-model	字符串	否	总结使用的 LLM 模型：gpt-4o/gpt-4o-mini（默认：gpt-4o-mini）
--preview	布尔	否	预览模式，不生成文件
--backup	布尔	否	启用备份（原始文件保存到 .ai_meeting_backup/）
--help	布尔	否	显示帮助信息

输出格式

Markdown（默认）

# 会议纪要

**日期**：2026-03-15  
**时长**：45分钟  
**参会人数**：5人

## 讨论要点

1. 讨论了Q1产品发布计划
2. 确定了市场营销策略
3. 分配了开发任务

## 行动项

- [ ] @张三：完成用户文档（3月20日前）
- [ ] @李四：设计海报素材（3月18日前）

## 决策点

- 采用A方案作为最终设计
- 预算增加10%

## 待办事项

- 下周一下午2点开会复盘

JSON

{
  "date": "2026-03-15",
  "duration": "45分钟",
  "participants": 5,
  "summary": "讨论了Q1产品发布计划...",
  "action_items": [
    {"assignee": "张三", "task": "完成用户文档", "due": "2026-03-20"}
  ],
  "decisions": ["采用A方案", "预算增加10%"],
  "todo": ["下周一下午2点开会复盘"]
}

技术栈

• 语音识别：openai-whisper-api（需 OPENAI_API_KEY）
• 文本处理：Python 标准库 + 正则
• LLM 总结：OpenAI GPT-4o-mini（默认）或 GPT-4o
• 文件处理：glob, json, pathlib

依赖

pip install openai python-dotenv

注意事项

• 需要有效的 OpenAI API key
• 音频文件支持：MP3, WAV, M4A, FLAC, OGG
• 较大文件处理时间较长（建议 < 30 分钟音频）
• 建议使用高质量录音以提高识别准确率

与其他技能配合

• audio-note-taker：类似功能，但 ai-meeting-helper 专为会议场景优化，输出更结构化
• wechat-formatter：生成的纪要可直接排版发布到公众号
• social-publisher：纪要内容可一键分发到多平台

License

MIT