Retrieval-Augmented Skill Optimization via Cross-Harness Adaptation

Jaewon Chu; Ji Soo Lee; Jihwan Park; Dohwan Ko; Jeehye Na; Seunghun Lee; Taehoon Lee; Minseo Yoon

cs.AISep 29, 2026

Retrieval-Augmented Skill Optimization via Cross-Harness Adaptation

Authors: Jaewon Chu, Ji Soo Lee, Jihwan Park, Dohwan Ko, Jeehye Na, Seunghun Lee, Taehoon Lee, Minseo Yoon, +3 more

Organizations: Korea University · KAIST · Meta AI

Abstract

An agent skill is a reusable, actionable natural-language artifact that guides an agent to perform a task effectively under a given harness. Recent studies have explored the optimization of agent skills, contributing to a growing collection of publicly available skills spanning diverse tasks, domains, and harnesses. Despite millions of publicly shared skills, existing skill optimization methods largely overlook this accumulated knowledge, instead relying solely on expensive agent rollouts to iteratively refine skills for a target task. To address this, we propose \textbf{Retrieval-Augmented Skill Optimization (RASO)}, a framework that leverages an external skill corpus as prior knowledge throughout skill optimization. RASO retrieves relevant knowledge from existing skills and adapts it to the target task and harness via Cross-Harness Adaptation, accounting for mismatches in both domain and harness. RASO comprises two complementary stages: \textbf{Retrieval-Augmented Skill Initialization (RASI)} constructs a knowledge-grounded initial skill without requiring agent rollouts, while \textbf{Retrieval-Augmented Skill Update (RASU)} iteratively refines the skill by retrieving external knowledge guided by execution feedback. Across four agent benchmarks and two models, extensive experiments show that RASO consistently outperforms baselines without retrieval-augmented skill initialization and updating.

Figures & tables

Figure 1: Comparison of RASO to prior skill optimization. Existing skill optimization methods (left) either generate an initial skill from the LLM alone or directly reuse retrieved skills, then refine them mainly from execution feedback. However, retrieved skills often come from mismatched harnesses; on SpreadsheetBench 92.9% originate from a different harness (right, top). RASO (middle) addresses this by retrieving relevant external skill knowledge and adapting it to the target domain and execution harness through Cross-Harness Adaptation , supporting both skill initialization (RASI) and skill update (RASU). Overall, our RASO results in substantially stronger performance than both direct retrieval and retrieval-free optimization methods.

Figure 2: Overview of RASO. RASI (top, without rollouts) generates retrieval queries from the task and harness descriptions, retrieves the top- $K$ skill sections per query from an external corpus, and adapts them into grounded lessons via Cross-Harness Adaptation to synthesize the initial skill $s_{0}$ . RASU (bottom, with rollouts) starts from $s_{0}$ , executes the current skill, and generates retrieval queries from rollout results. The retrieved knowledge is adapted into grounded lessons via Cross-Harness Adaptation and incorporated into skill.md to iteratively refine the skill. Each subsequent iteration performs new rollouts using the committed skill.

	Skill initialization (without rollout)
		Benchmark
Model	Method	OfficeQA	Spreadsheet	ALFWorld	WebShop
	No skill	11.44 $\pm$ 0.70	32.98 $\pm$ 0.12	64.43 $\pm$ 1.38	43.56 $\pm$ 1.01
	SkillRouter	11.44 $\pm$ 0.97	33.21 $\pm$ 0.94	55.97 $\pm$ 1.14	42.20 $\pm$ 0.37
	RFSI	40.11 $\pm$ 0.58	44.40 $\pm$ 2.74	69.40 $\pm$ 1.55	43.89 $\pm$ 1.17
	RASI	45.74 $\pm$ 1.40	49.17 $\pm$ 1.52	72.64 $\pm$ 1.63	45.06 $\pm$ 0.40

Table 1: Agent performance with GPT-5.6-Luna ( OpenAI, 2026 ) and Qwen-3.5-9B ( Qwen Team, 2026 ) . We separate skill initialization from skill update; initialization-only rows report performance at 0 rollouts. RFSI denotes retrieval-free skill initialization. Results are mean

\pm

standard error over three random seeds.

Table 4

Figure 4: Effect of skill corpus size with GPT-5.6-Luna. Dashed lines mark 0% (no corpus), and shading denotes standard error.

Figure 5: Qualitative example of Cross-Harness Adaptation. RASI retrieves a skill from a mismatched domain and harness (satellite imagery with numpy/raster arrays ) and adapts its procedure into the target spreadsheet harness. We compare the resulting lessons, initial skills, and generated code with (top) and without (bottom) the top-1 of the five retrieved sections given as input. Highlights mark the differences between the two runs, which are reflected in the generated code and its output. Without the top-1 skill, the model overlooks the missing group, causing an index shift.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Benchmark	Train	Validation	Test
OfficeQA	50	24	172
SpreadsheetBench	80	40	280
ALFWorld	39	18	134
WebShop	80	40	500

Appendix

Table 5: Number of tasks in each split.

Table 9

Method		Benchmark
Initialization	Update	OfficeQA	Spreadsheet
	$-$	40.11 $\pm$ 0.58	44.40 $\pm$ 2.74
	TextGrad	43.80 $\pm$ 1.27	49.64 $\pm$ 3.09
	GEPA	43.99 $\pm$ 1.40	54.53 $\pm$ 1.46
	SkillOpt	45.54 $\pm$ 2.05	57.02 $\pm$ 3.60
	WikiSkill	41.47 $\pm$ 1.18	47.14 $\pm$ 3.98

Appendix

Table 8: Update methods under a fixed RFSI initialization with GPT-5.6-Luna. The highlighted row is RASU.

Method	OfficeQA	Spreadsheet	ALFWorld	WebShop
TextGrad	1.45	0.97	2.04	41.03
GEPA	1.69	0.58	1.79	18.05
SkillOpt	2.23	2.24	2.75	17.86
WikiSkill	1.06	0.43	1.30	11.15
RASO	0.84	0.49	0.79	10.28

Appendix

Table 9: API cost (USD) of skill optimization with GPT-5.6-Luna. The lowest cost in each column is in bold.

Method	OfficeQA	Spreadsheet	ALFWorld	WebShop
TextGrad	196	320	114	360
GEPA	272	360	174	370
SkillOpt	236	360	154	400
WikiSkill	196	320	114	360
RASO	196	320	114	360

Appendix

Table 10: Number of rollouts used for skill optimization, excluding test-set evaluation.

Explore similar work

Skill-R1: Agent Skill Evolution via Reinforcement Learning
May 10, 2026Yash Vishe, Rohan Surana, Xunyi Jiang +8Self-Evolving Skill Libraries Unsupervised Skill Discovery
UC San Diego · Adobe Research
Skill Retrieval Augmentation for Agentic AI
Apr 27, 2026Weihang Su, Jianming Long, Qingyao Ai +6Skills Text Corpora
Department of Computer Science and Technology, Tsinghua University · ByteDance Inc.
COBRA-Skills: Contextual Bandit-Guided Evolution for Agent Skill Optimization
Sep 12, 2026Pingchen Lu, Xiangyi Wang, Xiang Li +6Self-Evolving Skill Libraries Agent Skill Retrieval
The Chinese University of Hong Kong, Shenzhen · Tianjin University · The Hong Kong University of Science and Technology (Guangzhou) +1

May 10, 2026, cs.LGJ/K move · Enter open · S save

Skill-R1: Agent Skill Evolution via Reinforcement Learning

Yash Vishe, Rohan Surana, Xunyi Jiang, Zihan Huang+7

UC San Diego · Adobe Research

Self-Evolving Skill Libraries Unsupervised Skill Discovery Skill Evolution Agentic Reinforcement Learning Skills

Open paper AskPDF arXiv

Agentic large language models often rely on skills, reusable natural language procedures that guide planning, action, and tool use. In practice, skills are typically improved through prompt engineering or by aligning the task LLM itself, which is costly, model-specific, and often infeasible for closed-source models. Skill optimization is not a one-step problem but a recurrent process with two coupled levels of credit assignment: a useful skill must improve rollout quality under current conditioning, while a useful revision must turn observed outcomes into a better skill for the next round. We propose Skill-R1, a reinforcement learning framework for instance-level recurrent skill optimization from verifiable rewards. Rather than updating the task LLM, Skill-R1 trains a lightweight skill generator that conditions on the task context, prior rollouts, and their verified outcomes to produce skills that steer a frozen task LLM. This preserves black-box compatibility with both open- and closed-source models while making adaptation substantially cheaper than model-level updates. Skill-R1 proceeds over multiple generations: at each step, the current skill induces rollouts whose verified outcomes are fed back to produce the next revision. To optimize this recurrent process, we introduce a bi-level group-relative policy optimization objective combining intra-generation and inter-generation advantages. The intra-generation term compares rollouts under shared skill conditioning, while the inter-generation term rewards revisions that improve behavior across successive generations. Together, these provide a principled objective for directional skill evolution rather than one-shot self-refinement. Empirically, Skill-R1 achieves consistent gains over no-skill baselines and standard GRPO across benchmarks with verifiable rewards, with particularly strong improvements on complex, multi-step tasks.

Abstract

Explore similar work

Skill-R1: Agent Skill Evolution via Reinforcement Learning

Skill Retrieval Augmentation for Agentic AI

COBRA-Skills: Contextual Bandit-Guided Evolution for Agent Skill Optimization