Agent Architecture Audit

Diagnose failures in agent systems by inspecting each layer that can change the model's behavior between the raw model call and the final user-visible output.

Contract

Inputs:

• Agent or LLM application repository, failing run, trace, log, or symptom
• Optional model/provider list, tool definitions, memory files, and UI output

Outputs:

• Severity-ranked findings with evidence references
• Layer-by-layer failure diagnosis
• Ordered fix plan that prefers code gates over prompt-only changes

Creates/Modifies: