@inproceedings{chen-etal-2026-repairgate,
title = "{R}epair{G}ate: {RL}-Trained Skill Repair and Abstention for Lifelong Agent Libraries",
author = "Chen, Xiaoyu and
Wu, Fengge and
Zhao, Junsuo and
Fan, Yun",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.648/",
doi = "10.18653/2026.findings-emnlp.648",
pages = "13610--13630",
ISBN = "979-8-89176-462-0",
abstract = "LLM agents increasingly rely on externalized skill libraries, yet existing maintainers typically delegate skill editing to a prompted LLM, leaving the question of how to train a skill maintainer open. We present RepairGate, a 9B drop-in wrapper around the standard candidate patcher: a counterfactually-trained post-trial gate accepts, blocks, or overrides each candidate patch, with a structured repair protocol for field-level edits and overlap-aware regularization to control library growth. Across six base agents on SkillFlow, RepairGate delivers +7.73 pp over Vanilla and +9.13 pp over prompted +Iter, with the largest gains on the weakest base agents where prompted iteration regresses, at a standardized $0.033/trial local overhead on top of the shared candidate-patching pipeline. Code and data: https://github.com/xxy33/RepairGate$"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-repairgate">
<titleInfo>
<title>RepairGate: RL-Trained Skill Repair and Abstention for Lifelong Agent Libraries</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xiaoyu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fengge</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junsuo</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yun</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>LLM agents increasingly rely on externalized skill libraries, yet existing maintainers typically delegate skill editing to a prompted LLM, leaving the question of how to train a skill maintainer open. We present RepairGate, a 9B drop-in wrapper around the standard candidate patcher: a counterfactually-trained post-trial gate accepts, blocks, or overrides each candidate patch, with a structured repair protocol for field-level edits and overlap-aware regularization to control library growth. Across six base agents on SkillFlow, RepairGate delivers +7.73 pp over Vanilla and +9.13 pp over prompted +Iter, with the largest gains on the weakest base agents where prompted iteration regresses, at a standardized 0.033/trial local overhead on top of the shared candidate-patching pipeline. Code and data: https://github.com/xxy33/RepairGate</abstract>
<identifier type="citekey">chen-etal-2026-repairgate</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.648</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.648/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>13610</start>
<end>13630</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T RepairGate: RL-Trained Skill Repair and Abstention for Lifelong Agent Libraries
%A Chen, Xiaoyu
%A Wu, Fengge
%A Zhao, Junsuo
%A Fan, Yun
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F chen-etal-2026-repairgate
%X LLM agents increasingly rely on externalized skill libraries, yet existing maintainers typically delegate skill editing to a prompted LLM, leaving the question of how to train a skill maintainer open. We present RepairGate, a 9B drop-in wrapper around the standard candidate patcher: a counterfactually-trained post-trial gate accepts, blocks, or overrides each candidate patch, with a structured repair protocol for field-level edits and overlap-aware regularization to control library growth. Across six base agents on SkillFlow, RepairGate delivers +7.73 pp over Vanilla and +9.13 pp over prompted +Iter, with the largest gains on the weakest base agents where prompted iteration regresses, at a standardized 0.033/trial local overhead on top of the shared candidate-patching pipeline. Code and data: https://github.com/xxy33/RepairGate
%R 10.18653/2026.findings-emnlp.648
%U https://aclanthology.org/2026.findings-emnlp.648/
%U https://doi.org/10.18653/2026.findings-emnlp.648
%P 13610-13630
Markdown (Informal)
[RepairGate: RL-Trained Skill Repair and Abstention for Lifelong Agent Libraries](https://aclanthology.org/2026.findings-emnlp.648/) (Chen et al., Findings 2026)
ACL