Problem
Communicate where model inference ends and application authorization begins.
System & architecture
Source data → Model / agent → Tool proposal → Policy / identity → Human approval → Audit
Data flow
Selecting an architecture layer reveals its responsibility and its security dependencies. No model calls or real actions are executed.
Technologies
TypeScript, React, semantic controls and SVG.
Engineering constraints
Architecture illustration only; not a deployed agent runtime or a security certification.
Testing
Keyboard selection, policy-boundary descriptions and alternate text are checked in browser QA.
Results & status
A public reference for discussing secure agent integration and human oversight.