Claude-First Infrastructure
Claude is powerful but opaque.
When you send a prompt to Claude, you have no visibility into what the model is doing, why it chose one approach over another, or how to course-correct when the output drifts.
The Solution
SubClaude is a shared infrastructure layer that sits in front of Claude. Instead of each team building their own proxy, SubClaude provides a common one that handles the work you would otherwise repeat: request logging, prompt versioning, parameter management, and model routing.
The core of SubClaude is a hooks system. Before a request goes to Claude, SubClaude can run a chain of hooks that inspect, transform, or reject the prompt.
How It Works
Claude Hooks
Hooks intercept requests and responses. Pre-hooks can inspect and transform prompts before they reach the model.
Common Proxy
A shared proxy layer that sits between your application and Claude.
Verdicts and Guidelines
Verdicts evaluate prompts against a set of rules or heuristics.
Hooks Library
A shared library of hooks that the community contributes to.
What You Get
Hooks
Write composable hooks that run before and after each Claude call.
Judges
Automated evaluators that score prompt quality against defined criteria.