linhut/gongwen-skill

This is a full-process processing tool for Chinese official government documents, built to comply with the national standard GB/T 9704 for official document formatting. It extracts the core formatting engine from the open-source desktop project AI Document Assistant, removes desktop and database dependencies, and is released as a standalone module. It supports all core official document workflows, from template generation, format checking and repair, content polishing, Markdown to docx conversion, to header, footer and page number injection, fact checking, and style enhancement. It natively integrates with the DeepSeek Harness (DSH) skill system and can be automatically discovered and loaded by DSH agents.

The tool targets users who need to write, revise, and format Chinese government official documents, and it adapts to multiple types of AI agent calling environments. For code-execution capable agents including DSH, Claude Code, and WorkBuddy, the tool can be called directly to read and write local files and complete document processing. For pure dialogue LLMs that do not support code execution, the agent can guide users to manually install and run the tool, then interpret the output results to help users complete their tasks.

The tool only depends on three pure Python packages: python-docx, pydantic, and pyyaml, and it requires no database or web framework to run. It is fully open-source under the MIT license, and by default it does not rely on third-party large language models, as all deterministic formatting capabilities are fully self-contained. If you need to enable LLM-assisted content optimization and internet fact checking, you can configure the corresponding environment variables to turn on these features. It also has built-in DNS pollution diagnostics that can automatically retry connections and give suggestions when GitHub access is abnormal in mainland China.