Analyze Project

When you need to understand a deep learning repository before touching anything, this is your read-only audit pass. It maps model structure, training and inference entrypoints, config relationships, and flags suspicious patterns without patching code or running heavy jobs. Think of it as reconnaissance before reproduction. The boundaries are clear: it won't debug tracebacks, set up environments, or speculatively refactor. You get structured markdown summaries of what's where, what looks risky, and where you'd likely need to hook in changes. Good for the moment when you clone an unfamiliar repo and need to know what you're actually looking at before committing to a reproduction attempt.