← 返回 Skills 市场
ebbfijsf

Agent Reader

作者 clare · GitHub ↗ · v1.3.7 · MIT-0
cross-platform ✓ 安全检测通过
392
总下载
1
收藏
2
当前安装
11
版本数
在 OpenClaw 中安装
/install agent-reader
功能描述
Document beautifier for AI Agents. Converts Markdown to styled webpages, Word, PDF, and image slideshows — the 'last mile' rendering engine for AI output. 专为...
使用说明 (SKILL.md)

Agent Reader

Document beautifier engine for AI Agents. Turn raw Markdown into delivery-ready documents. AI Agent 的文档美化引擎,把 Markdown 变成可交付的专业文档。

Tools / 工具

Tool What it does
render_markdown Markdown → styled webpage with sidebar TOC, code highlighting; can pre-generate PDF/DOCX files 带目录导航的网页,可预生成 PDF/DOCX
export_document Markdown → PDF or Word (.docx) with smart formatting 导出 PDF/Word
create_slideshow Images → full-screen slideshow with keyboard nav & auto-play 图片幻灯片
export_slideshow Slideshow → PDF or standalone HTML 导出幻灯片
open_file Smart open — auto-picks format based on user preferences 智能打开
configure_user_preferences Set default output format, theme, etc. 设置偏好
get_user_preferences Read current preferences 读取偏好

When to use / 什么时候调用

  • User wants a report formatted → export_document (format: "pdf" or "docx")
  • User says "make it a webpage" → render_markdown
  • User says "export to Word/PDF" → export_document
  • User has images for a presentation → create_slideshow or export_slideshow
  • User says "open this file" → open_file
  • No format specified → open_file (auto-selects based on preferences)

Setup / 接入

Add to your MCP config (Claude Desktop, Cline, OpenClaw, etc.):

{
  "mcpServers": {
    "agent-reader": {
      "command": "npx",
      "args": ["-y", "agent-reader", "mcp"]
    }
  }
}

Quick examples / 快速示例

Render a webpage 渲染网页

{ "content": "# My Report\
\
Hello world...", "theme": "light" }

Render with pre-export 预生成导出文件

{ "content": "# My Report\
\
Hello world...", "pre_export": ["pdf", "docx"] }

Export to PDF 导出 PDF

{ "content": "# My Report\
\
...", "format": "pdf" }

Export to Word 导出 Word

{ "content": "# My Report\
\
...", "format": "docx" }

Create slideshow from images 图片做幻灯片

{ "image_dir": "/path/to/images", "auto_play": 5 }

Key parameters / 关键参数

  • content — Markdown text to render/export Markdown 内容
  • format — "pdf" or "docx" (for export_document); "pdf" or "html" (for export_slideshow)
  • source_path — original .md file path, needed if Markdown references local images 原始文件路径(有本地图片时需要)
  • theme — "light" (default) or "dark"
  • pre_export — optional ["pdf"] or ["pdf", "docx"]; when returning a file path, MCP defaults to pre-generating PDF 可选预生成导出文件;MCP 返回文件路径时默认预生成 PDF
  • return_content — set true to get base64/HTML string instead of file path (for sandbox/Docker)

Notes / 注意事项

  • PDF export requires Puppeteer (auto-installed on first use) PDF 导出需要 Puppeteer
  • Word export works without Pandoc, but Pandoc produces better formatting Word 导出无需 Pandoc 也能用
  • return_content: true keeps inline HTML behavior unless pre_export is explicitly set return_content: true 默认保持内联返回;显式传 pre_export 时会落盘输出
  • All output writes to /tmp/agent-reader/ — no side effects 输出仅写入 /tmp
  • MIT licensed, open source 开源协议

Links

安全使用建议
This skill appears coherent for converting Markdown to webpages, PDFs, DOCX and slides. Before installing or enabling it: 1) verify the npm package and GitHub repo linked in SKILL.md (review package contents and maintainers) because the setup uses `npx` which fetches and runs remote code; 2) be aware Puppeteer will download Chromium (large binary and additional network activity); 3) avoid passing sensitive system paths as source_path or image_dir — the tool can read local files and can return file content (base64/HTML) which could be exfiltrated if misused; 4) prefer pinning an explicit package version instead of always pulling latest, and if possible run the MCP server for this skill in a sandboxed environment (container) to limit filesystem/network exposure. If you want, I can check the npm package or GitHub repo for more details before you install.
功能分析
Type: OpenClaw Skill Name: agent-reader Version: 1.3.7 The agent-reader skill is a document conversion utility designed to transform Markdown into PDF, DOCX, and HTML formats. The provided files (SKILL.md, schema.json) describe standard document processing functionality, with output restricted to /tmp/agent-reader/ and no evidence of data exfiltration, malicious execution, or prompt injection. While there is a discrepancy between the tools listed in the documentation and those defined in the schema (e.g., open_file is missing), this appears to be a development oversight rather than a malicious indicator.
能力评估
Purpose & Capability
Name, description, tools table and schema all describe Markdown → HTML/PDF/DOCX/slideshow conversion; required inputs (content, source_path, image_dir) and outputs align with the stated purpose. No unrelated credentials or binaries are requested.
Instruction Scope
SKILL.md stays within scope: it documents rendering/export APIs, mentions local source_path and image_dir (expected for images), and says output writes to /tmp/agent-reader. It does instruct adding an MCP config that uses `npx` to fetch/run the package — this causes remote code to be executed when installed, and the tool may read user-supplied local paths (source_path/image_dir), which is expected but should be used cautiously.
Install Mechanism
No install spec in the registry entry, but SKILL.md instructs using `npx -y agent-reader mcp` which will fetch the package from npm at runtime. The SKILL.md also warns Puppeteer will be auto-installed (and Puppeteer typically downloads a Chromium binary). These are normal for a rendering tool but do involve network downloads and running third-party code.
Credentials
The skill requests no environment variables or credentials. It accepts file paths (source_path, image_dir) which legitimately allow reading local files for conversion. Returning file content via base64/inline HTML is supported — this is expected but could leak sensitive data if misused.
Persistence & Privilege
Skill is not always-included, is user-invocable, and allows model invocation (normal). It does not request persistent system privileges or modify other skills' configurations.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install agent-reader
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agent-reader 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.3.7
Fix PDF button and strikethrough rendering
v1.3.6
Embed DOCX base64 in HTML for file:// download
v1.3.5
Fix file:// download CORS error
v1.3.4
Fix file:// download not triggering in Chrome
v1.3.3
Fix display name to Agent Reader
v1.3.2
Dramatically improved Word export quality: styled headings, tables with purple headers, blockquotes, code blocks with background. CLI and MCP now pre-generate PDF+DOCX by default. Fixed content loss and HTTP 500 bugs. Table cells center-aligned across all formats.
v1.3.1
Fix display name to Agent Reader
v1.3.0
Fix file:// protocol export buttons; add pre_export parameter for pre-generating PDF/DOCX alongside HTML
v1.2.2
- Updated skill version to 1.2.2. - SKILL.md now specifies MIT license and open source status in the Notes section. - Removed Glama AAA certification and "OpenClaw vetted" notices from documentation.
v1.2.1
Rewrite SKILL.md: bilingual (EN/CN), concise format, add security certifications (Glama AAA, OpenClaw vetted)
v1.2.0
- 新增详细功能列表和工具参数说明,介绍 Markdown 渲染、PDF/Word 导出、图片转幻灯片等能力 - 补充如何在 OpenClaw 接入,包括 MCP 直连与 CLI 调用示例 - 明确各工具的输入/输出字段及返回示例,方便集成和调试 - 标注常见错误类型和处理方法,提升易用性与可维护性 - 增加常用场景调用指引,让用户更易理解如何选择合适的工具
元数据
Slug agent-reader
版本 1.3.7
许可证 MIT-0
累计安装 2
当前安装数 2
历史版本数 11
常见问题

Agent Reader 是什么?

Document beautifier for AI Agents. Converts Markdown to styled webpages, Word, PDF, and image slideshows — the 'last mile' rendering engine for AI output. 专为... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 392 次。

如何安装 Agent Reader?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agent-reader」即可一键安装,无需额外配置。

Agent Reader 是免费的吗?

是的,Agent Reader 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Agent Reader 支持哪些平台?

Agent Reader 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Agent Reader?

由 clare(@ebbfijsf)开发并维护,当前版本 v1.3.7。

💬 留言讨论