PromptMN: Pseudo Prompting Language

A research team has proposed PromptMN, a domain-specific language designed to make human-to-AI instructions more explicit and less prone to misinterpretation, according to a paper posted to arXiv on June 15, 2026 [1]. The language, described as a "pseudo-prompting" tool, annotates natural language with compact, %-prefixed typed directives that specify roles, goals, requirements, priorities, constraints, plans, inputs, and outputs [1][2]. The authors argue that a significant portion of agent failures in software development and other workflows stem from context ambiguities rather than model limitations, as critical instructions are often buried in prose or left implicit [2]. PromptMN sits between informal prompting and programming-style pseudocode, making it structured enough to be inspectable and reusable while remaining accessible to analysts, managers, and developers across the software development lifecycle [2]. The system uses semantic resolution, which allows authors to write directives in any order while the model interprets them by function [2]. It also pairs with a technique called reverse prompt engineering, where a model restates a desired outcome in PromptMN syntax. This lets users inspect the inferred roles, goals, constraints, and missing assumptions before acting, which the paper says reduces repair cycles and yields a reusable artifact for aligning people and AI tools [2]. The feasibility study evaluated PromptMN across several frontier models: Claude Fable 5, Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5 [1][2]. Claude is a series of large language models developed by Anthropic, a San Francisco-based AI company founded in 2021 by former OpenAI members [6][8]. The models correctly resolved PromptMN instructions that included complex structures such as repetition, conditionals, methods, and a prime-checking task, all without fine-tuning [2]. The same vocabulary applied across new codebases, maintenance, and redesign scenarios presented in the study [2]. The paper acknowledges that large-scale validation remains future work, but states the early results suggest PromptMN is a practical step toward clearer, more reviewable human-to-AI interaction [1][2]. The research arrives as prompting has become the primary interface between humans and generative AI, yet natural language prompts remain fragile in production settings [2].