What it does
A research analysis of the Claude Code source code and how it is built. It argues that most of the codebase is infrastructure such as permission gates, context management, tool routing and recovery, rather than model logic. It also collects community analyses and a design-space guide for people building agent systems.
Use cases
- 01Study how permissions and tool routing are structured
- 02Compare design choices before building an agent harness
- 03Read the paper to brief a technical team