Prompt2Skill: Unsupervised Skill Optimization From Natural Language Instructions
The paper claims skills can be generated and optimized from a task description alone, without a curated training set.
Prompt2Skill turns a natural-language prompt into a task specification, then discovers or synthesizes data for iterative skill refinement. The loop uses reflective editing to tune the external skill artifact for the model that will consume it. The authors report gains across question answering, reading comprehension, spreadsheet manipulation, and math reasoning, with an average 10.8-point improvement over direct prompting. Source: HF Daily Papers' note
Prompt2Skill turns a natural-language prompt into a task specification, then discovers or synthesizes data for iterative skill refinement. The loop uses reflective editing to tune the external skill artifact for the model that will consume it. The authors report gains across question answering, reading comprehension, spreadsheet manipulation, and math reasoning, with an average 10.8-point improvement over direct prompting. Source: HF Daily Papers' note
score 4