Paper proposes PromptMN, a pseudo-prompting language to structure human-AI instructions

Prompting has become the primary interface between humans and generative AI, yet many natural language prompts remain fragile due to buried or implicit roles, goals, constraints, and expected outputs. In agentic and software development workflows, misreads at the first handoff can propagate through every step, with a significant portion of agent failures stemming from context ambiguities rather than model limitations.

PromptMN, introduced in a new arXiv paper, is a pseudo-prompting domain-specific language that annotates natural language with compact, %-prefixed typed directives. These directives cover roles, goals, requirements, priorities, constraints, plans, inputs, and outputs. Semantic resolution enables authors to write directives in any order, while the model interprets them by function, positioning PromptMN between informal prompting and programming-style pseudocode.

The language is designed to be structured enough for inspectability and reusability, yet lightweight enough for analysts, managers, developers, and stakeholders across the software development lifecycle (SDLC). It also pairs with reverse prompt engineering, where a model restates a desired outcome as PromptMN to surface inferred roles, goals, constraints, and missing assumptions before execution.

Feasibility of PromptMN was evaluated across several frontier models, including Claude Fable 5, Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5. The models correctly resolved PromptMN instructions, including complex structures such as repetition, conditionals, methods, and a prime-checking task, without fine-tuning. The same vocabulary applies across new codebases, maintenance, and redesign scenarios in SDLC contexts.

While large-scale validation remains future work, the authors suggest PromptMN is a practical step toward clearer, more reviewable human-to-AI interaction.