BREAKING
Claude Code Rebuilt From Scratch
0
components
0
core lines
0
stars
The Minimal Agent Loop
1
Call model
↓
2
Execute tools
↓
3
Repeat loop
It's the Harness, Not the Model
Harness Layer
focus
●
Permission governance
●
Context compression
●
Sub-agent teams
Model Layer
●
Anthropic, OpenAI
●
Groq, DeepSeek
●
Via LiteLLM proxy
Clean-Room After the Leak
Educational Value Drives Interest
AI NEWS BLITZ
A new Python project recreates Claude Code's architecture entirely from scratch.