[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-38916-en":3,"doc-seo-38916-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},38916,549758146520,"Patrick","https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470",8,"Research & Report","A Taxonomy of Failures in Tool-Augmented LLMs","Tool-augmented large language models (TALLMs) combine LLM reasoning with external tools such as APIs, search, and other learned or software components to extend capability in tasks beyond pure text prompting. Existing work emphasizes accuracy and capability improvements, while systematic characterization of failure modes and their root causes remains limited. This paper proposes a failure taxonomy for TALLMs, analyzes failures observed in two published systems (Gorilla and Chameleon), and provides recommendations for end-to-end testing, fault localization, and repair.","2025 IEEE/ACM International Conference on Automation of Software Test (AST) | 979-8-33 15-0 179-2/25/$31.00 ©2025 IEEE | DOI: 10. 1 109/AST66626 .2025.00019  \n2025 IEEE/ACM International Conference on Automation of Software Test (AST)  \nA Taxonomy of Failures in Tool-Augmented LLMs  \nCailin Winston University of Washington  \nSeattle, USA [cailinw@cs.washington.edu](cailinw@cs.washington.edu)  \nRen´e Just  \nUniversity of Washington Seattle, USA[rjust@cs.washington.edu](rjust@cs.washington.edu)  \nAbstract—Large language models (LLMs) can perform a variety of tasks given a user prompt that contains a description of the task. To enhance the performance of LLMs, recent research has focused on augmenting LLMs with external tools, such as Python APIs, REST APIs, and other deep learning models. Much of the research on tool-augmented LLMs (TALLMs) has focused on improving their capabilities and accuracy. However, research on understanding and characterizing the kinds of failures that can occur in these systems is lacking. To address this gap, this paper proposes a taxonomy of failures in TALLMs and their root causes, details an analysis of the failures that occur in two published TALLMs (Gorilla and Chameleon), and provides recommendations for testing and repair of TALLMs.  \nIndex Terms—Large language models, tool-augmented LLMs, software testing, fault localization, repair.  \nI. INTRODUCTION  \nLarge Language Models (LLMs) have demonstrated remarkable capabilities across a wide range of domains, performing tasks simply through text inputs that contain instructions and examples for the task. However, their effectiveness becomes limited in task domains that require nuanced reasoning, such as mathematics, or in more complex tasks that require interacting with external databases, APIs, or other software systems. To address these limitations, tool-augmented LLMs (TALLMs) extend the capabilities of LLMs by enabling them to use external tools, such as search engines, other deep learning models, modules for controlling physical agents, and domain-speci􀀂c functionality implemented in traditional software libraries. By enabling an LLM to utilize tools to acquire the most relevant and up-to-date information as well as interact with real-world systems, TALLMs can perform a wide range of tasks, such as question answering [1], embodied agent task planning [2], automatic program repair [3], and even customer-service chatbots [4] .  \nThe increased investment in research and deployment of TALLMs in real-world settings underscores the need for robust methods for testing and repair of TALLMs. Although prior work has developed benchmark datasets [5], [6] and failure detection methods [7], [8], systematic approaches to testing and repairing these systems are needed for a safe and wide-spread deployment of TALLMs. To achieve this goal, it is important to develop a comprehensive understanding of TALLM failure modes, which go beyond failures that can be localized to LLMs. Repair of TALLMs requires more than simply re-training an LLM, and thus mapping failures to root causes in these compositional systems is a crucial 􀀂rst step.  \nFig. 1: Common components of TALLMs.  \nThis paper focuses on developing a systematic understanding of failures in TALLMs. It enumerates possible failuresand their root causes, based on a literature survey and an in-depth analysis of two published TALLMs. Furthermore, it emphasizes the importance of end-to-end system testing and provides recommendations for leveraging fault localization and root cause analysis to evaluate, test, and repair TALLMs.  \nSpeci􀀂cally, this paper makes the following contributions:  \n• A systems and software engineering perspective on TALLMs in literature (Section II) .  \n• A systematic analysis of failures and root causes observed in two published TALLMs (Section III) .  \n• Recommendations for TALLM evaluation, testing, and repair (Section IV) .  \nThis paper aims to lay the foundation for enabling a more str","cbCaiibno066nNg9","https://ap.wps.com/l/cbCaiibno066nNg9","pdf",636724,4,1,11,"English","en",105,"# Introduction\n# A Systems and Software Engineering Perspective on TALLMs","[{\"question\":\"Why are tool-augmented LLMs needed beyond standard text-only prompting?\",\"answer\":\"TALLMs extend LLMs with external tools and system interactions, improving effectiveness in domains requiring nuanced reasoning and access to up-to-date or external information. They enable tasks that involve APIs, databases, or other software systems.\"},{\"question\":\"What gap does the paper address about research on TALLMs?\",\"answer\":\"The paper targets the lack of systematic understanding of the kinds of failures that occur in TALLMs and the root causes behind those failures. It goes beyond focusing only on failures attributable to the LLM itself.\"},{\"question\":\"How does the paper recommend evaluating and repairing TALLMs?\",\"answer\":\"It emphasizes end-to-end system testing and recommends using fault localization and root cause analysis to evaluate, test, and repair TALLMs. Repair is framed as mapping failures to causes in compositional system components, not only retraining the LLM.\"}]",1783075159,28,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"a-taxonomy-of-failures-in-tool-augmented-llms","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":20},"https://docshare.wps.com/document/a-taxonomy-of-failures-in-tool-augmented-llms/38916/",{"url":52,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-09","2026-07-03",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why are tool-augmented LLMs needed beyond standard text-only prompting?","Question",{"text":75,"@type":76},"TALLMs extend LLMs with external tools and system interactions, improving effectiveness in domains requiring nuanced reasoning and access to up-to-date or external information. They enable tasks that involve APIs, databases, or other software systems.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What gap does the paper address about research on TALLMs?",{"text":80,"@type":76},"The paper targets the lack of systematic understanding of the kinds of failures that occur in TALLMs and the root causes behind those failures. It goes beyond focusing only on failures attributable to the LLM itself.",{"name":82,"@type":73,"acceptedAnswer":83},"How does the paper recommend evaluating and repairing TALLMs?",{"text":84,"@type":76},"It emphasizes end-to-end system testing and recommends using fault localization and root cause analysis to evaluate, test, and repair TALLMs. Repair is framed as mapping failures to causes in compositional system components, not only retraining the LLM.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]