267 lines
7.1 KiB
Markdown
Executable File
267 lines
7.1 KiB
Markdown
Executable File
---
|
||
created: 2026-05-04
|
||
type: guide
|
||
tags: [Claude Code, Obsidian, 数据导出, 对话存档, SQLite]
|
||
---
|
||
|
||
# Claude Code 对话导出到 Obsidian 完整方案
|
||
|
||
> 将所有 Claude Code 对话会话存档到 Obsidian 知识库
|
||
> 归档时间：2026-05-04
|
||
|
||
---
|
||
|
||
## 📌 原理
|
||
|
||
Claude Code 的对话数据存储在本地 SQLite 数据库中：
|
||
|
||
**数据库路径**：
|
||
- macOS/Linux：`~/.claude/CLAUDE.md` 同级目录下的 `~/.claude/conversations.db`
|
||
- 实际数据库文件：`~/.claude/*.db` 或 `~/.claude/PROJECT_ID/*.db`
|
||
|
||
**数据库结构**（典型）：
|
||
- `conversations` 表：会话元数据（ID、标题、创建时间等）
|
||
- `messages` 表：对话消息（角色、内容、时间戳）
|
||
- `attachments` 表：附件/文件引用
|
||
|
||
---
|
||
|
||
## 🔧 方案一：Python 脚本导出（推荐）
|
||
|
||
### 1. 导出脚本
|
||
|
||
```python
|
||
#!/usr/bin/env python3
|
||
"""
|
||
claude_to_obsidian.py
|
||
将 Claude Code 所有对话导出为 Obsidian Markdown 笔记
|
||
"""
|
||
|
||
import sqlite3
|
||
import os
|
||
import json
|
||
from datetime import datetime
|
||
from pathlib import Path
|
||
|
||
# 配置
|
||
CLAUDE_DB_PATH = os.path.expanduser("~/.claude/conversations.db")
|
||
OBSIDIAN_PATH = "/obsidian/ClaudeCode对话/" # 你的 Obsidian vault 路径
|
||
|
||
def export_all():
|
||
if not os.path.exists(CLAUDE_DB_PATH):
|
||
print(f"❌ 数据库不存在: {CLAUDE_DB_PATH}")
|
||
print("请确认 Claude Code 已安装并运行过")
|
||
return
|
||
|
||
os.makedirs(OBSIDIAN_PATH, exist_ok=True)
|
||
|
||
conn = sqlite3.connect(CLAUDE_DB_PATH)
|
||
conn.row_factory = sqlite3.Row
|
||
cursor = conn.cursor()
|
||
|
||
# 获取所有会话
|
||
cursor.execute("""
|
||
SELECT c.id, c.title, c.created_at, c.updated_at,
|
||
COUNT(m.id) as msg_count
|
||
FROM conversations c
|
||
LEFT JOIN messages m ON c.id = m.conversation_id
|
||
GROUP BY c.id
|
||
ORDER BY c.updated_at DESC
|
||
""")
|
||
|
||
conversations = cursor.fetchall()
|
||
print(f"📊 找到 {len(conversations)} 个会话")
|
||
|
||
exported = 0
|
||
for conv in conversations:
|
||
title = conv['title'] or f"untitled-{conv['id'][:8]}"
|
||
# 文件名安全处理
|
||
safe_title = "".join(c for c in title if c not in r'\/:*?"<>|')
|
||
safe_title = safe_title[:80] # 限制长度
|
||
filename = f"{safe_title}.md"
|
||
filepath = os.path.join(OBSIDIAN_PATH, filename)
|
||
|
||
# 获取该会话的所有消息
|
||
cursor.execute("""
|
||
SELECT role, content, created_at
|
||
FROM messages
|
||
WHERE conversation_id = ?
|
||
ORDER BY created_at ASC
|
||
""", (conv['id'],))
|
||
|
||
messages = cursor.fetchall()
|
||
|
||
# 生成 Markdown
|
||
md = generate_markdown(conv, messages)
|
||
|
||
with open(filepath, 'w', encoding='utf-8') as f:
|
||
f.write(md)
|
||
|
||
exported += 1
|
||
print(f" ✅ {title} ({len(messages)} 条消息)")
|
||
|
||
print(f"\n🎉 完成！导出 {exported} 个会话到 {OBSIDIAN_PATH}")
|
||
conn.close()
|
||
|
||
def generate_markdown(conv, messages):
|
||
created = datetime.fromtimestamp(conv['created_at']).strftime('%Y-%m-%d %H:%M:%S') if conv['created_at'] else 'unknown'
|
||
updated = datetime.fromtimestamp(conv['updated_at']).strftime('%Y-%m-%d %H:%M:%S') if conv['updated_at'] else 'unknown'
|
||
|
||
md = f"""---
|
||
created: {created}
|
||
updated: {updated}
|
||
tags: [claude-code, 对话]
|
||
conversation_id: {conv['id']}
|
||
message_count: {len(messages)}
|
||
---
|
||
|
||
# {conv['title'] or 'Untitled'}
|
||
|
||
> 创建时间：{created}
|
||
> 更新时间：{updated}
|
||
|
||
---
|
||
|
||
## 对话记录
|
||
|
||
"""
|
||
|
||
for msg in messages:
|
||
role = msg['role'] # 'user' or 'assistant'
|
||
content = msg['content'] or ''
|
||
ts = datetime.fromtimestamp(msg['created_at']).strftime('%H:%M:%S') if msg['created_at'] else ''
|
||
|
||
if role == 'user':
|
||
md += f"\n### 👤 用户 `{ts}`\n\n"
|
||
else:
|
||
md += f"\n### 🤖 Claude `{ts}`\n\n"
|
||
|
||
# 处理内容（可能是 JSON 字符串）
|
||
md += sanitize_content(content)
|
||
md += "\n\n---\n"
|
||
|
||
return md
|
||
|
||
def sanitize_content(content):
|
||
"""清理内容，移除可能的工具调用元数据"""
|
||
try:
|
||
# 尝试解析 JSON
|
||
data = json.loads(content)
|
||
if isinstance(data, dict):
|
||
# 提取文本内容
|
||
if 'content' in data:
|
||
return data['content']
|
||
if 'text' in data:
|
||
return data['text']
|
||
return json.dumps(data, indent=2, ensure_ascii=False)
|
||
except (json.JSONDecodeError, TypeError):
|
||
pass
|
||
|
||
return str(content)
|
||
|
||
if __name__ == "__main__":
|
||
export_all()
|
||
```
|
||
|
||
### 2. 使用
|
||
|
||
```bash
|
||
# 安装依赖（通常不需要）
|
||
pip install sqlite3 # Python 内置
|
||
|
||
# 运行导出
|
||
python3 claude_to_obsidian.py
|
||
```
|
||
|
||
---
|
||
|
||
## 🔧 方案二：直接查看数据库结构
|
||
|
||
```bash
|
||
# 查看数据库表
|
||
sqlite3 ~/.claude/conversations.db ".tables"
|
||
|
||
# 查看表结构
|
||
sqlite3 ~/.claude/conversations.db ".schema conversations"
|
||
sqlite3 ~/.claude/conversations.db ".schema messages"
|
||
|
||
# 查看会话数量
|
||
sqlite3 ~/.claude/conversations.db "SELECT COUNT(*) FROM conversations;"
|
||
|
||
# 导出单个会话为 JSON
|
||
sqlite3 -json ~/.claude/conversations.db \
|
||
"SELECT m.* FROM messages m JOIN conversations c ON m.conversation_id = c.id WHERE c.title LIKE '%关键词%';"
|
||
```
|
||
|
||
---
|
||
|
||
## 🔧 方案三：按项目/目录组织
|
||
|
||
Claude Code 为每个项目创建独立的数据库：
|
||
|
||
```bash
|
||
# 查找所有项目数据库
|
||
find ~/.claude -name "*.db" -type f
|
||
|
||
# 输出示例：
|
||
# ~/.claude/projects/project-uuid-1/conversations.db
|
||
# ~/.claude/projects/project-uuid-2/conversations.db
|
||
```
|
||
|
||
**按项目导出脚本**：
|
||
|
||
```python
|
||
import os
|
||
import glob
|
||
|
||
def export_by_project():
|
||
db_files = glob.glob(os.path.expanduser("~/.claude/**/conversations.db"), recursive=True)
|
||
|
||
for db_path in db_files:
|
||
# 从路径提取项目名
|
||
project_name = db_path.split(os.sep)[-2] # 倒数第二级是项目 UUID
|
||
project_dir = os.path.join(OBSIDIAN_PATH, project_name)
|
||
os.makedirs(project_dir, exist_ok=True)
|
||
|
||
print(f"📁 处理项目: {project_name}")
|
||
# ... 使用上面的导出逻辑
|
||
```
|
||
|
||
---
|
||
|
||
## 📂 Obsidian 目录结构建议
|
||
|
||
```
|
||
/obsidian/ClaudeCode对话/
|
||
├── _templates/
|
||
│ └── Claude对话模板.md # Obsidian 模板
|
||
├── 2026-05-04/
|
||
│ ├── 修复API鉴权Bug.md
|
||
│ ├── 编写单元测试.md
|
||
│ └── 重构数据库模块.md
|
||
└── INDEX.md # 索引笔记（按日期/项目分类）
|
||
```
|
||
|
||
---
|
||
|
||
## ⚠️ 注意事项
|
||
|
||
1. **隐私安全**：对话可能包含敏感信息（API Key、代码等），导出后注意权限控制
|
||
2. **大文件**：长对话可能生成大文件，Obsidian 打开会慢
|
||
3. **编码问题**：确保以 UTF-8 编码写入
|
||
4. **增量导出**：记录上次导出的时间戳，只导出新对话
|
||
5. **工具调用**：Claude Code 的工具调用（文件读写、Shell 执行）元数据可能需要特殊处理
|
||
|
||
---
|
||
|
||
## 🔄 自动同步方案
|
||
|
||
```bash
|
||
# 加入 cron，每天凌晨自动导出
|
||
0 2 * * * cd /path/to/scripts && python3 claude_to_obsidian.py >> /tmp/claude_export.log 2>&1
|
||
```
|
||
|
||
---
|
||
|
||
*研究归档，2026-05-04 | Claude Code 对话导出到 Obsidian 完整指南*
|