antvis/chart-visualization-skills

What it solves

This project provides a set of specialized skills and tools to help AI agents and LLMs generate accurate, production-ready data visualizations. It bridges the gap between natural language requests and the complex specifications required by the AntV visualization libraries (G2, G6, X6), reducing common coding errors and migration pitfalls.

How it works

The project implements a "Harness Engineering" approach to improve LLM accuracy. It offers several components:

• AI Skills: A collection of specialized generators for different visualization needs, including G2 (statistical charts), G6 (graph/network analysis), X6 (interactive editors), GPT-Vis (AI-native syntax), and infographic creators.
• Context Retrieval Service: An HTTP API, CLI, and TypeScript SDK that allows AI models to retrieve the most relevant documentation, constraints, and code snippets (using hybrid search) to ensure the generated code is correct.
• Integration: The skills can be added to AI coding tools like Claude Code or installed via npx for multiple agents.

Who it’s for

Developers and data analysts who use AI agents to automate the creation of charts, graphs, and infographics, as well as those integrating AntV libraries into AI-powered data analysis workflows.

Highlights

• High Accuracy: Rigorously tested against 174 cases, achieving up to 98.2% accuracy for G2 charts.
• Comprehensive Coverage: Supports over 26 chart types, 10+ layout algorithms for graphs, and 50+ infographic templates.
• Hybrid Retrieval: Uses a combination of Full-Text Search (FTS) and Vector (HNSW ANN) search with RRF fusion for precise context retrieval.
• AI-Native Syntax: Includes GPT-Vis, a framework-agnostic, markdown-like syntax designed specifically for LLM generation.