用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/javimosch/open-claw-skills --skill deshell命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | deshell |
| description | Fetch web pages as clean Markdown and search the web via the DeShell proxy |
| version | 1.5.0 |
| metadata | {"openclaw":{"emoji":"🔍","requires":{"bins":"[Truncated]","env":"[Truncated]"},"primaryEnv":"DESHELL_API_KEY","install":["[Truncated]"]}} |
Gives agents discoverable, consistent access to the DeShell proxy — no manual URL construction, no remembering headers or API keys.
DeShell converts web pages into clean Markdown, saving 60–80% of tokens for LLM consumption.
Get your API key from https://deshell.ai (sign up or use your existing key)
Install the deshell CLI manually (one-time):
npm install @deshell/mcp
Set the DESHELL_API_KEY environment variable
# Fetch any URL as clean Markdown
deshell fetch https://example.com
# Search the web and get results as Markdown
deshell search "best practices for Go error handling"
# Multi-word queries work naturally — no quoting needed
deshell search top 10 AI companies 2025
# Take a screenshot of a web page and return it as an image
deshell screenshot https://example.com
# Render a web page (such as a single page javascript app) before trying to extract markdown
deshell render https://example.com
# Fetch a URL and return its raw content bypassing any attempt to render markdown
deshell raw https://example.com
# Fetch a URL and return its content without using the cache
deshell nocache https://example.com
| Environment Variable | Default | Description |
|---|---|---|
DESHELL_API_KEY | (none) | API key |
DESHELL_PROXY_URL | https://proxy.deshell.ai/ | Proxy base URL |
DESHELL_EXTRA_HEADERS | (none) | Comma-separated extra headers in Header-Name:value format |
deshell fetch — returns page content as Markdown on stdoutdeshell search — returns search results with titles, URLs, descriptions, and page content as Markdown on stdoutTo pass additional DeShell headers, use comma-separated Header-Name:value pairs:
# Single header
DESHELL_EXTRA_HEADERS="X-DeShell-No-Cache:true" deshell fetch https://example.com
# Multiple headers
DESHELL_EXTRA_HEADERS="X-DeShell-No-Cache:true,X-DeShell-Max-Tokens:2000" deshell fetch https://example.com
# Research a topic
deshell search "OpenClaw agent framework"
# Read documentation
deshell fetch https://docs.github.com/en/rest
# Force fresh fetch (bypass cache)
DESHELL_EXTRA_HEADERS="X-DeShell-No-Cache:true" deshell fetch https://news.ycombinator.com
If you prefer not to install npm packages, you can call the proxy directly with curl:
# Fetch a page
curl -s "https://proxy.deshell.ai/https://example.com" \
-H "X-DeShell-Key: YOUR_API_KEY"
# Search the web
curl -s "https://proxy.deshell.ai/search?q=your+query" \
-H "X-DeShell-Key: YOUR_API_KEY" \
-H "Accept: text/markdown"