# Skill Optimization

Prompt2Skill is a multi-agent framework that, from a natural-language task description alone, has a Data Agent derive a task specification and build a proxy data pool through retrieval, adaptation, and synthesis, and a Skill Agent refine the skill in a closed loop of execution feedback, reflective editing, and paired statistical screening, improving over direct prompting by 36.1% and over off-the-shelf skills by 83.0% on average across question answering, reading comprehension, spreadsheet manipulation, and mathematical reasoning with five open-source and commercial target models, without significant regressions.

Prompt2Skill is a multi-agent framework that, from a natural-language task description alone, has a Data Agent derive a task specification and build a proxy data pool through retrieval, adaptation, and synthesis, and a Skill Agent refine the skill in a closed loop of execution feedback, reflective editing, and paired statistical screening, improving over direct prompting by 36.1% and over off-the-shelf skills by 83.0% on average across question answering, reading comprehension, spreadsheet manipulation, and mathematical reasoning with five open-source and commercial target models, without significant regressions.

The work proposes Retrieval-Augmented Skill Optimization (RASO), a framework that retrieves section-level knowledge from a public skill corpus and adapts it to the target task and harness through Cross-Harness Adaptation, where RASI builds an initial skill without any agent rollouts and RASU iteratively refines the skill by retrieving missing knowledge guided by execution feedback; across four benchmarks (OfficeQA, SpreadsheetBench, ALFWorld, WebShop) and two models (GPT-5.6-Luna, Qwen-3.5-9B), RASO consistently outperforms baselines without retrieval-augmented initialization and updating.

The work proposes Retrieval-Augmented Skill Optimization (RASO), a framework that retrieves section-level knowledge from a public skill corpus and adapts it to the target task and harness through Cross-Harness Adaptation, where RASI builds an initial skill without any agent rollouts and RASU iteratively refines the skill by retrieving missing knowledge guided by execution feedback; across four benchmarks (OfficeQA, SpreadsheetBench, ALFWorld, WebShop) and two models (GPT-5.6-Luna, Qwen-3.5-9B), RASO consistently outperforms baselines without retrieval-augmented initialization and updating.

The work represents LLM agent skills as graph-structured natural-language artifacts, where each node is an execution step with its operational guidance and directed edges encode context-dependent transitions between steps, and builds GraphSkillEvo, a population-based evolutionary framework with global-guidance mutation, graph-structure mutation, global-guidance crossover, and graph-structure crossover that searches this structured skill space, outperforming the skill-optimization baseline SkillOpt on average across five agent benchmarks, two LLMs, and two execution settings (+4.01% on GPT-5.4-nano and +1.76% on GPT-5.4) while consuming fewer total optimization tokens.

The work represents LLM agent skills as graph-structured natural-language artifacts, where each node is an execution step with its operational guidance and directed edges encode context-dependent transitions between steps, and builds GraphSkillEvo, a population-based evolutionary framework with global-guidance mutation, graph-structure mutation, global-guidance crossover, and graph-structure crossover that searches this structured skill space, outperforming the skill-optimization baseline SkillOpt on average across five agent benchmarks, two LLMs, and two execution settings (+4.01% on GPT-5.4-nano and +1.76% on GPT-5.4) while consuming fewer total optimization tokens.