README.md

metadata

title: Agent Architecture Visualizer
emoji: 🔄
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 4.36.0
app_file: app.py
pinned: false

Agent Architecture Visualizer

Interactive visualization of Claude Code's agent loop architecture based on the paper "Dive into Claude Code: The Design Space of Today's and Future AI Agent Systems" (arXiv:2604.14228).

Core Concepts Demonstrated

1. The Agent Loop: Model → Tools → Repeat cycle with state persistence
2. Permission Gates: 7-mode permission system with ML-based classification
3. Context Compaction: Five-layer pipeline for managing context windows
4. Subagent Delegation: Worktree-isolated task spawning

Paper Insight

The paper reveals that while the core agent loop is simple (a while-loop calling the model and executing tools), most complexity lives in the surrounding systems: permission classification, context management, extensibility mechanisms, and session storage.

Space Features

• Interactive agent loop simulator
• Permission mode visualizer
• Context compaction demo
• Subagent delegation trace

Citation

@article{liu2026claudecode,
  title={Dive into Claude Code: The Design Space of Today's and Future AI Agent Systems},
  author={Liu, Jiacheng and Zhao, Xiaohan and Shang, Xinyi and Shen, Zhiqiang},
  year={2026}
}