Genie Training the Wisher: Six-Dimension Task-Agnostic AI Coaching for Learning Transferable LLM Prompting Skills

Effective prompting has become a key skill for interacting with Large Language Models (LLMs), yet most existing support focuses on optimizing prompts for specific tasks or domains. In this work, we introduce Prompting Coach (PC), a task-agnostic prompting support tool designed to help users improve their general prompting skills, rather than performance on a single prompt or application. Grounded in a conceptual framework that defines six core dimensions of prompting, our tool captures prompt qualities and information types that recur across diverse prompting patterns and interaction paradigms. PC provides heuristic feedback on user prompts, highlighting strengths and weaknesses along these dimensions and offering actionable suggestions without relying on task-specific templates or domain knowledge. To evaluate its effectiveness, we conducted a controlled user study in which participants () interacted with PC under different configurations and across sequences of heterogeneous tasks. We measured both task outcomes and prompt quality over time. Users who received prompting feedback significantly improved their task performance and prompt quality compared to the baseline condition. Importantly, these improvements generalized beyond the immediate task: participants who used PC on earlier tasks demonstrated higher performance when approaching new, previously unseen tasks, even when feedback was absent. These findings suggest that PC, grounded in six prompting dimensions, supports transferable learning of prompting strategies rather than short-term optimization. Overall, while LLM users may improve through repeated unsupported use, automated task-agnostic support accelerates the development of transferable prompting skills. This provides insights for the design of AI literacy programs and further understanding of prompting as a learnable, generalizable interaction skill.