arXiv: Structured Intent as Protocol-Like Communication Layer — Cross-Model Robustness for Agent Interoperability

arXiv·low signal

Peng Gang investigates how reliably structured intent representations preserve user goals across different AI models, languages, and prompting frameworks. The paper evaluates PPS (Prompt Protocol Specification) as a protocol-like communication layer and identifies a 'weak-model compensation effect' where structured intent representations disproportionately improve performance on weaker models. Relevant to multi-agent systems where orchestrator agents must reliably communicate intent to heterogeneous specialist agents across different model providers.