PromptMN proposes typed directives for less ambiguous AI prompts

A new arXiv paper introduces PromptMN, a pseudo-prompting language for reducing ambiguity in AI prompts. It adds compact typed directives for roles, goals, requirements, priorities, constraints, plans, inputs, and outputs.

The idea targets a common failure mode in agentic workflows: important instructions are buried in prose or interpreted inconsistently at the first handoff. Those misunderstandings can then propagate through the whole task.

PromptMN keeps natural language while adding structure, aiming to make prompts easier for models to parse and for humans to audit.