Markdown.new Skill
低风险
作者:joelchance | 审计时间:2026-02-26T09:59:20.936Z | 规则集:0.2.0
技能介绍
Use this skill to convert public URLs into LLM-ready Markdown via markdown.new.
✨ Resolve relative paths like
scripts/... and references/... from the skill directory, not workspace root. ✨ If current directory is unknown, use an absolute script path.
✨
auto: default. Let markdown.new use its fastest successful pipeline. ✨
ai: force Workers AI HTML-to-Markdown conversion. ✨
browser: force headless browser rendering for JS-heavy pages. ✨ Use
--deliver-md to force file output in .md format. 使用场景
文档(原文)
来源:README.md 以下为作者原文(通常为英文)。安装请以页面顶部“快速安装”为准。
Markdown.new Skill
Single-skill repository for markdown-new - official Cloudflare URL-to-Markdown service (markdown.new) converted into a skill.
Skill entrypoint:
markdown-new/SKILL.md
What It Does
markdown-new converts public web pages into LLM-ready Markdown using markdown.new, with:
- URL-to-Markdown conversion for summarization, extraction, RAG, and archiving
- conversion fallback control (
auto,ai,browser) - optional image retention
- optional wrapped delivery mode for downstream parsing
Path Resolution (Important)
- Relative paths such as
scripts/markdown_new_fetch.pyare relative to the skill directory. - Do not run
python3 scripts/markdown_new_fetch.py ...from workspace root unlessscripts/exists there. - Safe command from any current directory:
python3 ~/.codex/skills/markdown-new/scripts/markdown_new_fetch.py 'https://example.com'
Modes
Conversion Modes (--method)
auto: default pipeline, fastest successful pathai: force Workers AI conversion pathbrowser: force Browser Rendering for JS-heavy pages
Output Modes
- default: print Markdown to stdout
--output <file>: write Markdown to file--deliver-md: write.mdoutput with wrapped content; useful for reasoning LLMs on long reads because it reduces format confusion:
<url>
...markdown...
</url>
If --deliver-md is used without --output, filename is auto-generated from the URL.
How It Works
- Validate the input URL (
http/https). - Call
POST https://markdown.new/withurl,method, andretain_images. - Accept response as either raw markdown or JSON with markdown in
content. - Normalize metadata and choose output behavior.
- Return stdout by default,
--outputfor files, and--deliver-mdfor wrapped.mdpackets.
Install Paths
- Codex (macOS/Linux):
~/.codex/skills/markdown-new - Claude Code (macOS/Linux):
~/.claude/skills/markdown-new
Install on macOS/Linux (single command)
Codex
mkdir -p ~/.codex/skills && rm -rf ~/.codex/skills/markdown-new && cp -R /Users/pro16/Dropbox/experiments/skills-i-use/markdown-new ~/.codex/skills/
Claude Code
mkdir -p ~/.claude/skills && rm -rf ~/.claude/skills/markdown-new && cp -R /Users/pro16/Dropbox/experiments/skills-i-use/markdown-new ~/.claude/skills/
Quick Usage
python3 scripts/markdown_new_fetch.py 'https://example.com'
python3 scripts/markdown_new_fetch.py 'https://example.com' --method browser --retain-images --output page.md
python3 scripts/markdown_new_fetch.py 'https://example.com' --deliver-md
Credits
webservervisfor the markdown conversion service powering this skill.
安全审计
低风险
摘要
Convert public web pages into clean Markdown with markdown.new for AI workflows. Use when tasks require URL-to-Markdown conversion for summarization, RAG ing...
风险画像
ToxicSkills 分析
黑名单
未命中
提示词注入
未检测到
Toxic 标签
exfiltration
当前静态检测未发现 Toxic 信号。
关键风险 0 项
暂无 LLM 风险要点(LLM 未启用或无缓存)。
确定性发现(证据)
| 规则 | 严重性 | 文件 | 片段 |
|---|---|---|---|
| NET_HTTP_REQUEST | 中 | skills/joelchance/markdown-convert/scripts/markdown_new_fetch.py 行 10 | import urllib.request |
| NET_HTTP_REQUEST | 中 | skills/joelchance/markdown-convert/scripts/markdown_new_fetch.py 行 71 | def build_request(api_url: str, payload: Dict[str, object]) -> urllib.request.Request: |
| NET_HTTP_REQUEST | 中 | skills/joelchance/markdown-convert/scripts/markdown_new_fetch.py 行 73 | return urllib.request.Request( |
| NET_HTTP_REQUEST | 中 | skills/joelchance/markdown-convert/scripts/markdown_new_fetch.py 行 185 | with urllib.request.urlopen(req, timeout=args.timeout) as resp: |
| QUALITY_README_PRESENT | 低 | README 行 无 | README detected |
评分标准
每个技能从 5 个维度评分,加权总分决定星级。
代码毒性 100/100 (权重 30%)
隐私风险 52/100 (权重 25%)
权限范围 90/100 (权重 20%)
作者声誉 75/100 (权重 15%)
代码质量 78/100 (权重 10%)
星级说明
5★ 安全 — 总分 ≥ 80
4★ 良好 — 总分 70–79
3★ 注意 — 总分 60–69
2★ 有风险 — 总分 40–59
1★ 危险 — 总分 < 40
为何是这个评分?
以下维度评分低于 60 分,拉低了整体评级:
- 隐私风险: 52/100