forked from Selig/openclaw-skill
6 custom skills (assign-task, dispatch-webhook, daily-briefing, task-capture, qmd-brain, tts-voice) with technical documentation. Compatible with Claude Code, OpenClaw, Codex CLI, and OpenCode.
66 lines
1.8 KiB
Markdown
Executable File
66 lines
1.8 KiB
Markdown
Executable File
# Skill Seeker Patches
|
||
|
||
## fetch_and_save 工具
|
||
|
||
這個 patch 為 Skill Seeker MCP 新增 `fetch_and_save` 工具,可以直接抓取網頁並存檔,**不經過 Claude API**,節省 97%+ 的 token 成本。
|
||
|
||
### 安裝步驟
|
||
|
||
```bash
|
||
# 1. 確保已安裝 skill-seekers
|
||
pip install skill-seekers
|
||
|
||
# 2. 執行 patch 腳本
|
||
python add_fetch_and_save.py
|
||
|
||
# 3. 重啟 Claude Code
|
||
```
|
||
|
||
### 使用方式
|
||
|
||
#### 單檔模式
|
||
```
|
||
mcp__skill-seeker__fetch_and_save(
|
||
url="https://docs.example.com/guide.md",
|
||
output="docs/guide.md"
|
||
)
|
||
```
|
||
|
||
#### 批次模式
|
||
```
|
||
mcp__skill-seeker__fetch_and_save(urls=[
|
||
{"url": "https://docs.example.com/intro.md", "output": "docs/intro.md"},
|
||
{"url": "https://docs.example.com/api.md", "output": "docs/api.md"}
|
||
])
|
||
```
|
||
|
||
### 參數說明
|
||
|
||
| 參數 | 類型 | 預設值 | 說明 |
|
||
|------|------|--------|------|
|
||
| `url` | str | - | 單一 URL(需搭配 output) |
|
||
| `output` | str | - | 輸出檔案路徑 |
|
||
| `urls` | list | - | 批次模式:`[{"url": str, "output": str}, ...]` |
|
||
| `extract_markdown` | bool | true | 自動從 HTML 提取 markdown |
|
||
| `timeout` | int | 30 | 請求超時(秒) |
|
||
| `rate_limit` | float | 0.5 | 請求間隔(秒) |
|
||
|
||
### Token 效率比較
|
||
|
||
| 方法 | Token 消耗 | 說明 |
|
||
|------|-----------|------|
|
||
| WebFetch + Write | ~2000/頁 | 內容經過 API 兩次 |
|
||
| **fetch_and_save** | **~50/頁** | 只有呼叫和狀態回傳 |
|
||
|
||
### 如何提示 LLM 使用這個工具
|
||
|
||
> "請用 `fetch_and_save` 工具抓取網頁並存檔,這樣比較省 token"
|
||
|
||
或者在 system prompt 中加入:
|
||
> "當需要抓取網頁存檔時,優先使用 `mcp__skill-seeker__fetch_and_save` 而非 WebFetch,以節省 token"
|
||
|
||
### 注意事項
|
||
|
||
- 每次 `pip upgrade skill-seekers` 後需要重新執行 patch
|
||
- Patch 腳本會檢查是否已 patch,不會重複修改
|