alchaincyf/huashu-report

What it solves

It solves the problem of AI agents producing generic, poorly structured reports that lack the professional polish and rigor of top-tier institutional research. Instead of simply aggregating search results into a document, it provides a framework for agents to produce industry reports, white papers, and academic papers that follow the actual design and writing conventions used by leading global organizations.

How it works

The project is implemented as an "Agent Skill" consisting of text-based guidelines and Python utility files. It reverse-engineers the patterns found in 41+ high-quality reports from institutions like Stanford HAI, McKinsey, and OpenAI. When triggered, the agent adopts four distinct roles—researcher, editor, information designer, and data visualization specialist—to execute a production pipeline. It includes a library of SVG charts, a rendering engine for automatic table-of-contents generation, and a set of strict guidelines on layout, color usage, and writing style (e.g., writing conclusions in chart titles rather than topics).

Who it’s for

It is designed for users of AI agents (such as Claude Code or other custom agent platforms) who need to generate professional-grade research reports, white papers, data insights, or arXiv-style papers.

Highlights

• Evidence-Based Standards: Conventions are derived from a quantitative analysis of 41 real-world institutional reports.
• Role-Based Workflow: Moves beyond simple text generation by simulating a multi-role professional editorial process.
• Visual Tooling: Includes a Python-based inline SVG library for 8 different chart types and a CSS-based rendering system.
• Strict Editorial Logic: Provides a clear decision criterion for content inclusion: if removing a sentence doesn't change the reader's judgment of the conclusion, it is deleted as "work log."