Raw LLM-driven KiCad agents cannot be trusted to generate board and schematic files directly, because the file formats contain implicit relationships the model does not understand. The safe architecture is: the model proposes intent, deterministic tools execute, and a verification engine produces evidence.
From the 2026-10-04 pain-radar digest (finding 2): a design-proposal doc written explicitly because raw LLM agents fail at direct file generation. Handle-only author and a proposal rather than an incident, so Low confidence. Included because it names the structural gap behind several other entries: no deterministic execution and verification between the agent and the file.