@inproceedings{zhang-etal-2026-agent-skills,
title = "From Agent Skills to Executable Environments: Frontier-Targeted Training for Tool-Use {LLM}s",
author = "Zhang, Dawei and
Huangfu, Yubin and
Yao, Kyle Yusong and
Wei, Boyu and
Han, Zongqi and
Wu, Likang and
Zhao, Hongke and
He, Ming and
Fan, Jianping",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.305/",
pages = "7026--7050",
ISBN = "979-8-89176-459-0",
abstract = "LLMs increasingly use external tools and stateful environments for complex, multi-step tasks. Further improving their tool-use capabilities, however, remains bottlenecked by the lack of scalable stateful environments and verified multi-turn trajectories matched to the target LLM{'}s current capability boundary. We observe that agent skills, packages pairing workflow specifications with executable scripts, provide scalable seed material for tool-use environment synthesis. This motivates Grafting, a skill-to-environment synthesis loop in which coding-agent harnesses perform open-ended construction under deterministic validation contracts, producing self-contained executable environments that are refined using trajectory-level verdicts. In an initial end-to-end run, Grafting produces 500 environments and approximately 12K verified multi-turn trajectories. We train on these trajectories with Fathoming, an iterative GRPO procedure that selects verified trajectories near the target model{'}s capability boundary. Training a 4B-parameter instruction-tuned model with two rounds of Fathoming on the trajectory pool produces Fathom-4B and yields strong out-of-distribution gains: BFCLv4 multi-turn accuracy rises from 22.0 to 39.7 (+17.7 points), and ComplexFuncBench per-call accuracy rises from 49.6 to 61.6 (+12.0 points). Using the same environment-synthesis pipeline, we further construct Grove-5K, a collection of 5,101 executable environments. We release Grafting, Grove-500, Grove-5K, and Fathom-12K."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-agent-skills">
<titleInfo>
<title>From Agent Skills to Executable Environments: Frontier-Targeted Training for Tool-Use LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dawei</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yubin</namePart>
<namePart type="family">Huangfu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kyle</namePart>
<namePart type="given">Yusong</namePart>
<namePart type="family">Yao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Boyu</namePart>
<namePart type="family">Wei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zongqi</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Likang</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongke</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ming</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianping</namePart>
<namePart type="family">Fan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>LLMs increasingly use external tools and stateful environments for complex, multi-step tasks. Further improving their tool-use capabilities, however, remains bottlenecked by the lack of scalable stateful environments and verified multi-turn trajectories matched to the target LLM’s current capability boundary. We observe that agent skills, packages pairing workflow specifications with executable scripts, provide scalable seed material for tool-use environment synthesis. This motivates Grafting, a skill-to-environment synthesis loop in which coding-agent harnesses perform open-ended construction under deterministic validation contracts, producing self-contained executable environments that are refined using trajectory-level verdicts. In an initial end-to-end run, Grafting produces 500 environments and approximately 12K verified multi-turn trajectories. We train on these trajectories with Fathoming, an iterative GRPO procedure that selects verified trajectories near the target model’s capability boundary. Training a 4B-parameter instruction-tuned model with two rounds of Fathoming on the trajectory pool produces Fathom-4B and yields strong out-of-distribution gains: BFCLv4 multi-turn accuracy rises from 22.0 to 39.7 (+17.7 points), and ComplexFuncBench per-call accuracy rises from 49.6 to 61.6 (+12.0 points). Using the same environment-synthesis pipeline, we further construct Grove-5K, a collection of 5,101 executable environments. We release Grafting, Grove-500, Grove-5K, and Fathom-12K.</abstract>
<identifier type="citekey">zhang-etal-2026-agent-skills</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.305/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>7026</start>
<end>7050</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T From Agent Skills to Executable Environments: Frontier-Targeted Training for Tool-Use LLMs
%A Zhang, Dawei
%A Huangfu, Yubin
%A Yao, Kyle Yusong
%A Wei, Boyu
%A Han, Zongqi
%A Wu, Likang
%A Zhao, Hongke
%A He, Ming
%A Fan, Jianping
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F zhang-etal-2026-agent-skills
%X LLMs increasingly use external tools and stateful environments for complex, multi-step tasks. Further improving their tool-use capabilities, however, remains bottlenecked by the lack of scalable stateful environments and verified multi-turn trajectories matched to the target LLM’s current capability boundary. We observe that agent skills, packages pairing workflow specifications with executable scripts, provide scalable seed material for tool-use environment synthesis. This motivates Grafting, a skill-to-environment synthesis loop in which coding-agent harnesses perform open-ended construction under deterministic validation contracts, producing self-contained executable environments that are refined using trajectory-level verdicts. In an initial end-to-end run, Grafting produces 500 environments and approximately 12K verified multi-turn trajectories. We train on these trajectories with Fathoming, an iterative GRPO procedure that selects verified trajectories near the target model’s capability boundary. Training a 4B-parameter instruction-tuned model with two rounds of Fathoming on the trajectory pool produces Fathom-4B and yields strong out-of-distribution gains: BFCLv4 multi-turn accuracy rises from 22.0 to 39.7 (+17.7 points), and ComplexFuncBench per-call accuracy rises from 49.6 to 61.6 (+12.0 points). Using the same environment-synthesis pipeline, we further construct Grove-5K, a collection of 5,101 executable environments. We release Grafting, Grove-500, Grove-5K, and Fathom-12K.
%U https://aclanthology.org/2026.emnlp-main.305/
%P 7026-7050
Markdown (Informal)
[From Agent Skills to Executable Environments: Frontier-Targeted Training for Tool-Use LLMs](https://aclanthology.org/2026.emnlp-main.305/) (Zhang et al., EMNLP 2026)
ACL
- Dawei Zhang, Yubin Huangfu, Kyle Yusong Yao, Boyu Wei, Zongqi Han, Likang Wu, Hongke Zhao, Ming He, and Jianping Fan. 2026. From Agent Skills to Executable Environments: Frontier-Targeted Training for Tool-Use LLMs. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 7026–7050, Budapest, Hungary. Association for Computational Linguistics.