Skill
meeting-ingest
stanley6635/meeting-ingest
51
InicialPluginScore · de 0 a 100Claude Code / OpenCode skill: process voice-to-text meeting transcripts into structured wiki knowledge
Instalação
$
npx convez-ai stanley6635/meeting-ingest1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Estrelas
1Forks
0Acompanhando
0Saúde
42%Último commit
há 4 mesesVersão
—Issues abertas
0Contribuidores
1README
A tradução para o português deste README ainda está na fila.
meeting-ingest · 会议转录入库
中文
把语音转文字的会议记录,自动纠正错误、对照知识库去重、提炼真正有价值的信息,写入 wiki。
它解决什么问题
语音转录的会议记录通常有三个痛点:
- 转写错误太多——人名、产品名、术语被转得面目全非
- 90% 是已知信息——会上说的东西,wiki 里早就记过了
- 执行细节混在战略判断里——操作指引、截止日期、现场情绪充斥全文
meeting-ingest 用一个 7 步管线解决这三个问题:
flowchart LR
A["📝 转录文件"] --> B["Step 0\n归档"]
B --> C["Step 1\n纠错 pass"]
C --> D["Step 2\n搜索 wiki"]
D --> E["Step 3\n读完整页面"]
E --> F["Step 4\n三道筛子过滤"]
F --> G["Step 5\n报摘要"]
G --> H["Step 6\n写回 wiki"]
H --> I["Step 7\nLint 检查"]
style A fill:#1F497D,color:#fff
style I fill:#4BACC6,color:#fff
核心能力
- 强制纠错:逐字对照 wiki 中已记录的人名、产品名、术语,把转写错误揪出来
- 全文对比:搜索 wiki 后必须读完匹配页面的完整内容,不能只看搜索片段就判"已知"
- 三道筛子:
- 已知 vs 新 —— wiki 已有的,剔除
- 一周测试 —— 一周后还重要吗?
- 持久层 vs 执行层 —— 一次 60 分钟会议,真正入库的通常 ≤3 个要点
快速开始
git clone https://github.com/stanley6635/meeting-ingest.git ~/.claude/skills/meeting-ingest
然后编辑 skill.md 里的配置部分,设置你的知识库路径。详见 SETUP.md。
依赖
- Python 3(stdlib only,无需 pip install)
- 结构化 wiki 知识库(包含
people/、products/、mechanisms/等子目录) file-ingestskill(可选,用于自动归档)
English
A Claude Code / OpenCode skill that processes voice-to-text meeting transcripts into structured wiki knowledge.
Pipeline
flowchart LR
A["📝 Transcript"] --> B["Step 0\nArchive"]
B --> C["Step 1\nError Fix"]
C --> D["Step 2\nWiki Search"]
D --> E["Step 3\nRead Full Pages"]
E --> F["Step 4\nThree Filters"]
F --> G["Step 5\nSummarize"]
G --> H["Step 6\nWrite-back"]
H --> I["Step 7\nLint"]
style A fill:#1F497D,color:#fff
style I fill:#4BACC6,color:#fff
Key Features
| Step | What | Why |
|---|---|---|
| 1. Error Correction | Fix ASR mangling of names, products, terms | Garbage in → garbage out |
| 2. Wiki Search | Cross-reference every topic domain | Don't re-record what you already know |
| 3. Full Page Read | Read matched pages completely | Snippet match ≠ already covered |
| 4. Three Filters | Known/new, one-week test, durable vs execution | 90% of meeting content doesn't belong in long-term memory |
| 5-6. Write-back | Structured updates with time layers | Preserve history, don't flatten |
| 7. Lint | Verify source paths and cross-links | No fabricated references |
Quick Start
git clone https://github.com/stanley6635/meeting-ingest.git ~/.claude/skills/meeting-ingest
Edit the 配置 section in skill.md to match your knowledge base paths. See SETUP.md for details.
Requirements
- Python 3 (stdlib only)
- Structured wiki knowledge base
file-ingestskill (optional)
License
MIT