document-parser

Extract structured data from PDFs, images, and Word files with layout analysis, table recognition, OCR, seal detection, and directory extraction.

Security Scan

The skill implements a document parsing client that behaves as advertised, but by default it uploads user files to an unknown IP-based API endpoint (47.111.146.164) without requiring credentials — which is disproportionate and risky unless you deliberately configure a trusted server.

Purpose & Capability

The code and documentation consistently implement a remote-document-parser client (PDF/image/Word parsing, OCR, table and seal detection). Using a remote API for heavy tasks like OCR/layout analysis is reasonable, so the capability aligns with the name/description. However, the packaged default base_url is an IP address (47.111.146.164) embedded in examples and defaults, which is unexpected for a generic skill and should be justified by the author.

Instruction Scope

Runtime instructions and the CLI cause the skill to read local files and POST their binary contents to a remote HTTP endpoint. The SKILL.md and config examples explicitly point to the same unknown IP. The skill will attempt uploads even without an API key (it logs a warning but proceeds), so users could inadvertently exfiltrate sensitive documents simply by running the default parse command.

Install Mechanism

This is instruction-only plus a Python script; there is no download-from-URL or post-install arbitrary code fetch. Dependencies are standard (requests, python-docx, Pillow) and listed in requirements.txt. No high-risk install behavior was found.

Credentials

The skill does not require environment variables, but supports optional DOCUMENT_PARSER_API_KEY and DOCUMENT_PARSER_BASE_URL. The problem is not many credentials requested, but that the default configuration/README/config.example hardcodes an explicit IP-based endpoint. Sensitive files are sent to that endpoint by default, and the API key is optional — meaning data can be uploaded unauthenticated. That is disproportionate for a drop-in skill where users may expect local processing or to configure their own server.

Persistence & Privilege

The package does not request always:true, does not modify other skills or system-wide settings, and only writes output files derived from user input to the current working directory. It does read a local config.json if present (expected). No elevated persistence or privilege escalation behavior observed.

What to consider before installing

This skill is functionally coherent but risky by default: if you run 'document-parser parse <file>' as-is, the file will be uploaded to the default server at 47.111.146.164 (the README and config example point to that IP). Before installing/using it, consider: 1) Do not upload sensitive documents to an unknown third party. 2) Prefer to set DOCUMENT_PARSER_BASE_URL to a trusted/self-hosted parser endpoint, or host the parsing service yourself. 3) If you must use this skill, require and provide an API key and confirm the operator/trustworthiness of the endpoint. 4) Audit network traffic (or run in an isolated environment) to verify where files are sent. 5) If you don't have a trusted remote parser, avoid using the skill or inspect/modify index.py to implement local processing instead.

高精度文档解析技能，从 PDF、图片、Word 文档中提取结构化数据。

用途

• 解析 PDF、图片 (JPG/PNG)、Word 文档
• 版面分析与结构提取
• 表格识别（输出 HTML/Markdown）
• OCR 文字识别
• 印章检测
• 目录提取

命令

解析文档

document-parser parse <文件路径> [选项]

示例：

document-parser parse C:\docs\report.pdf
document-parser parse C:\docs\scan.jpg --layout --table
document-parser parse C:\docs\contract.docx --output markdown

查询任务状态

document-parser status <任务 ID>

参数说明

参数	说明	示例
文件路径	PDF/图片/Word 文件路径	C:\docs\report.pdf
--layout	启用版面分析	--layout
--table	启用表格识别	--table
--seal	启用印章检测	--seal
--output	输出格式 (json/markdown/both)	--output markdown
--pages	页码范围	--pages 1-5,8,10-12

配置

方式一：环境变量

DOCUMENT_PARSER_API_KEY=your_api_key
DOCUMENT_PARSER_BASE_URL=http://47.111.146.164:8088/taidp/v1/idp/general_parse

方式二：配置文件

在技能目录创建 config.json：

{
  "api_key": "your_api_key",
  "base_url": "http://47.111.146.164:8088/taidp/v1/idp/general_parse"
}

输出格式

返回结构化 JSON 包含：

• pages: 解析后的页面数组
• elements: 版面元素（文本、表格、图片等）
• markdown: Markdown 格式文本
• data: 数据统计摘要

依赖

• requests
• python-docx (Word 支持)
• Pillow (图片处理)

错误码

错误码	消息	说明
10000	Success	识别成功
10001	Missing parameter	参数缺失
10002	Invalid parameter	非法参数
10003	Invalid file	文件格式非法
10004	Failed to recognize	识别失败
10005	Internal error	内部错误