pith. machine review for the scientific record. sign in

arxiv: 2604.08906 · v1 · submitted 2026-04-10 · 💻 cs.SE

Recognition: no theorem link

Dissecting Bug Triggers and Failure Modes in Modern Agentic Frameworks: An Empirical Study

Xiaowen Zhang , Hannuo Zhang , Shin Hwei Tan
Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:44 UTC · model grok-4.3

classification 💻 cs.SE
keywords agentic frameworksbug analysisempirical studymulti-agent systemsorchestration faultsfailure modesLLM reliability
0
0 comments X

The pith

Analysis of 409 fixed bugs shows agentic frameworks have distinct failure modes from autonomous orchestration.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper analyzes 409 fixed bugs collected from five representative agentic frameworks to map out failure modes that arise specifically from their autonomous multi-agent design. It introduces a five-layer abstraction that ranges from high-level orchestration down to infrastructure layers in order to classify symptoms and causes systematically. The work identifies symptoms such as unexpected execution sequences and ignored user configurations that do not appear in earlier pipeline-style LLM tools, along with root causes centered on model faults, context mismanagement, and orchestration errors. Statistical checks confirm these patterns hold consistently across the frameworks and that certain trigger combinations recur. Readers would care because these findings supply concrete targets for testing and repair in systems that are increasingly deployed for independent task execution.

Core claim

The study establishes that agentic frameworks exhibit specialized symptoms unique to autonomous orchestration, including unexpected execution sequences and ignored user configurations, together with agent-specific root causes such as model-related faults, cognitive context mismanagement, and orchestration faults. These dimensions display cross-framework consistency, measurable statistical associations, and recurring triggering patterns such as model backend combinations that transfer across different framework designs.

What carries the argument

A five-layer abstraction spanning orchestration to infrastructure that classifies structural complexities and organizes bug symptoms and root causes within agentic frameworks.

If this is right

  • Cross-framework consistency supports the creation of shared testing suites that target orchestration and model layers.
  • Frequent patterns such as model backend-ID combinations can be encoded into automated detectors usable in multiple frameworks.
  • Identified associations between symptoms and root causes allow developers to prioritize fixes in context management and orchestration logic.
  • Transferability of patterns across designs reduces the need for framework-specific bug studies.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Tool builders could embed the five-layer model into runtime monitors that flag orchestration anomalies before they reach production.
  • The observed consistency suggests that training data for agent debugging models could be pooled across frameworks without major loss of applicability.
  • Extending the pattern mining to live execution traces rather than only fixed bugs would test whether the same triggers appear in unfixed field failures.

Load-bearing premise

The 409 fixed bugs drawn from the five chosen frameworks stand in for the full spectrum of problems in agentic systems and the five-layer abstraction accounts for their structural features.

What would settle it

A new agentic framework whose fixed bugs largely fail to match the reported symptoms, root-cause categories, or five-layer structure would disprove the generality of the findings.

Figures

Figures reproduced from arXiv: 2604.08906 by Hannuo Zhang, Shin Hwei Tan, Xiaowen Zhang.

Figure 1
Figure 1. Figure 1: Conceptual architecture of agentic frameworks. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Simplified code snippets illustrating representative root causes. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Jensen-Shannon similarity between frameworks. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Simplified triggering scenarios with labels. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
read the original abstract

Modern agentic frameworks (e.g., CrewAI and AutoGen) have evolved into complex, autonomous multi-agent systems, introducing unique reliability challenges beyond earlier pipeline-based LLM libraries. However, existing empirical studies focus on earlier LLM libraries or task-level bugs, leaving the unique complexities of these agentic frameworks unexplored. We bridge the gap by conducting a comprehensive study of 409 fixed bugs from five representative agentic frameworks. We propose a five-layer abstraction to capture structural complexities in agentic frameworks, spanning from orchestration to infrastructure. Our study uncovers specialized symptoms, such as unexpected execution sequences and user configurations ignored, which are unique to autonomous orchestration. We further identify agent-specific root causes, including modelrelated faults, cognitive context mismanagement, and orchestration faults. Statistical analysis reveals cross-framework consistency and significant associations among these bug dimensions. Finally, our automated pattern mining identifies frequent bug-triggering patterns (e.g., model backend-ID combinations), and we show their transferability across different framework designs. Our findings facilitate cross-platform testing and improve the reliability of agentic systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

4 major / 2 minor

Summary. The paper conducts an empirical study of 409 fixed bugs drawn from five representative agentic frameworks (e.g., CrewAI, AutoGen). It introduces a five-layer abstraction spanning orchestration to infrastructure, identifies specialized symptoms unique to autonomous orchestration (unexpected execution sequences, ignored user configurations) and agent-specific root causes (model-related faults, cognitive context mismanagement, orchestration faults), reports cross-framework statistical consistency and associations, and mines transferable bug-triggering patterns via automated analysis.

Significance. If the sampling and classification hold, the work would offer timely empirical grounding for reliability challenges in autonomous multi-agent systems, distinguishing them from earlier pipeline-style LLM libraries. The dataset size and cross-framework pattern transferability findings are concrete strengths that could inform testing practices.

major comments (4)
  1. [§3] §3 (Bug Collection and Sampling): The paper does not specify the exact criteria or search queries used to identify the 409 fixed bugs, nor does it address selection bias inherent in studying only resolved issues. Fixed bugs form a filtered subset that may under-represent persistent or low-reproducibility failures, directly undermining the claim that observed symptoms and root causes are representative of modern agentic systems.
  2. [§4.1] §4.1 (Framework Selection): No explicit justification or coverage analysis is given for choosing the five frameworks as representative. Without documented selection criteria (e.g., diversity in memory models or multi-agent topologies), the reported cross-framework consistency and statistical associations risk being artifacts of the chosen corpus rather than intrinsic properties.
  3. [§4.2] §4.2 (Five-Layer Abstraction): The five-layer model is presented as capturing structural complexities, yet the manuscript provides no derivation process, inter-rater validation, or completeness check against the collected bugs. This assumption is load-bearing for attributing symptoms and root causes to specific layers.
  4. [§5] §5 (Classification and Statistical Analysis): Details on bug classification criteria, inter-rater agreement metrics, and the exact statistical tests (including p-values and effect sizes) used to establish associations and consistency are missing. These omissions leave the central claims about specialized symptoms and agent-specific causes only partially supported.
minor comments (2)
  1. [Abstract] Abstract: 'modelrelated' should be hyphenated as 'model-related'.
  2. [§5] The manuscript would benefit from a table summarizing the distribution of bugs across the five frameworks and layers to improve readability of the quantitative results.

Simulated Author's Rebuttal

4 responses · 0 unresolved

We appreciate the referee's constructive feedback on our empirical study of bugs in agentic frameworks. We address each major comment in detail below, agreeing to enhance the manuscript with additional methodological details and clarifications to improve transparency and support for our claims.

read point-by-point responses

  1. Referee: [§3] §3 (Bug Collection and Sampling): The paper does not specify the exact criteria or search queries used to identify the 409 fixed bugs, nor does it address selection bias inherent in studying only resolved issues. Fixed bugs form a filtered subset that may under-represent persistent or low-reproducibility failures, directly undermining the claim that observed symptoms and root causes are representative of modern agentic systems.

    Authors: We thank the referee for highlighting this gap. In the revised manuscript, we will explicitly document the search queries and criteria used to collect the 409 fixed bugs from the GitHub repositories of the five frameworks. We will also add a discussion on selection bias, explaining that while fixed bugs provide confirmed cases with developer-validated resolutions, they may not capture all failure modes. This limitation will be noted, but we argue that the patterns identified remain valuable for understanding prevalent issues in these systems. revision: yes

  2. Referee: [§4.1] §4.1 (Framework Selection): No explicit justification or coverage analysis is given for choosing the five frameworks as representative. Without documented selection criteria (e.g., diversity in memory models or multi-agent topologies), the reported cross-framework consistency and statistical associations risk being artifacts of the chosen corpus rather than intrinsic properties.

    Authors: We concur that the framework selection process should be better justified. The updated version will include a clear description of the selection criteria, emphasizing factors such as GitHub popularity, community adoption, and architectural diversity (e.g., variations in agent topologies and memory mechanisms). We will also provide a brief coverage analysis to demonstrate that the chosen frameworks are representative and that the consistency findings are likely generalizable. revision: yes

  3. Referee: [§4.2] §4.2 (Five-Layer Abstraction): The five-layer model is presented as capturing structural complexities, yet the manuscript provides no derivation process, inter-rater validation, or completeness check against the collected bugs. This assumption is load-bearing for attributing symptoms and root causes to specific layers.

    Authors: The five-layer abstraction was iteratively developed from the frameworks' source code and documentation to systematically categorize the structural elements involved in bug manifestation. For the revision, we will detail this derivation process, report inter-rater agreement statistics for the layer assignments, and include a completeness verification showing that the layers adequately cover all observed bugs in our dataset. revision: yes

  4. Referee: [§5] §5 (Classification and Statistical Analysis): Details on bug classification criteria, inter-rater agreement metrics, and the exact statistical tests (including p-values and effect sizes) used to establish associations and consistency are missing. These omissions leave the central claims about specialized symptoms and agent-specific causes only partially supported.

    Authors: We will revise §5 to include the detailed classification guidelines used for symptoms and root causes. Additionally, we will report inter-rater reliability measures, such as Cohen's kappa, and specify the statistical tests applied (including p-values and effect sizes) to substantiate the associations and cross-framework consistency. These additions will provide the necessary rigor to support our key findings. revision: yes

Circularity Check

0 steps flagged

Purely empirical bug study with no derivations, predictions, or self-referential reductions

full rationale

This is an empirical study that collects 409 fixed bugs from five frameworks, proposes a five-layer abstraction to organize observed symptoms and root causes, and performs statistical analysis on the data. No equations, fitted parameters, or mathematical derivations appear in the abstract or described methodology. The central claims about specialized symptoms (e.g., unexpected execution sequences) and agent-specific root causes rest on direct inspection of the bug corpus rather than any self-definition, fitted-input prediction, or load-bearing self-citation chain. The five-layer model is presented as a descriptive abstraction derived from the data, not as a result forced by prior self-cited theorems or ansatzes. Representativeness of the sample is an external validity concern, not a circularity issue in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The claims rest on assumptions about the representativeness and accuracy of fixed-bug data plus the validity of the newly proposed abstraction; no free parameters or invented physical entities are involved.

axioms (2)
  • domain assumption Fixed bugs reported in open-source repositories accurately reflect the true failure modes and root causes of the frameworks.
    The study infers root causes directly from developer-provided fixes and bug reports.
  • domain assumption The five chosen frameworks are representative of the broader class of modern agentic frameworks.
    Findings are generalized from this sample to agentic systems as a whole.
invented entities (1)
  • Five-layer abstraction no independent evidence
    purpose: To capture structural complexities in agentic frameworks spanning orchestration to infrastructure.
    This model is introduced by the authors to organize their bug analysis.

pith-pipeline@v0.9.0 · 5486 in / 1295 out tokens · 42608 ms · 2026-05-10T17:44:29.153567+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

56 extracted references · 17 canonical work pages · 2 internal anchors

  1. [1]

    2023.Add improved sources splitting in BaseQA WithSourcesChain.https://github .com/langchain-ai/langchain/pull/8716

    2023

  2. [2]

    2024.[BUG] Flow @listen with _and error.https://github.com/crewAIInc/crewA I/issues/1463

    2024

  3. [3]

    2024.Exceptions in Orleans storage.https://github.com/microsoft/autogen/issue s/4490

    2024

  4. [4]

    https://www.ilsilfve rskiold.com/articles/agentic-aI-comparing-new-open-source-frameworks

    2025.Agentic AI: Comparing New Open-Source Frameworks. https://www.ilsilfve rskiold.com/articles/agentic-aI-comparing-new-open-source-frameworks

    2025

  5. [5]

    2025.AzureAISearchTool Not working as expected.https://github.com/microsoft /autogen/issues/6308

    2025

  6. [6]

    2025.[BUG] Authentication Error When Using OpenAI Compatible LLMs.https: //github.com/crewAIInc/crewAI/issues/2647

    2025

  7. [7]

    2025.[BUG] Error when using ToolCallingAgent as manager.https://github.com /huggingface/smolagents/issues/606

    2025

  8. [8]

    2025.[Bug] Final answer is too greedy in ToolCallingAgent parallel calls.https: //github.com/huggingface/smolagents/issues/1481

    2025

  9. [9]

    2025.Bug: GraphFlow with termination condition automatically ends after first query.https://github.com/microsoft/autogen/issues/6746

    2025

  10. [10]

    2025.[BUG] Tool validation with executor type defined.https://github.com/huggi ngface/smolagents/issues/913

    2025

  11. [11]

    2025.Could not send Multiple system message at autogen.https://github.com/mic rosoft/autogen/issues/6116

    2025

  12. [12]

    https://github.com/crewAIInc/crewAI

    2025.CrewAI. https://github.com/crewAIInc/crewAI

    2025

  13. [13]

    2025.Customized ChatAgentContainer not support.https://github.com/microsoft /autogen/issues/6730

    2025

  14. [14]

    2025.It seems that num_gpu does not work with ollama models.https://github.c om/langchain-ai/langchain/issues/32059

    2025

  15. [15]

    https://github.com/langchain-ai/langchain

    2025.LangChain. https://github.com/langchain-ai/langchain

    2025

  16. [16]

    2025.LangChain Indexing API: Incorrect num_skipped Count Due to Missing Within-Batch Deduplication Tracking.https://github.com/langchain-ai/langcha in/issues/32272

    2025

  17. [17]

    https://github.com/langchain-ai/langgraph

    2025.LangGraph. https://github.com/langchain-ai/langgraph

    2025

  18. [18]

    2025.LLMCallEvent fails to log “tools” to be chosen from by the LLM in BaseOpe- nAIChatCompletionClient.https://github.com/microsoft/autogen/issues/6531

    2025

  19. [19]

    2025.LocalCommandLineCodeExecutor to support PowerShell.https://github.com /microsoft/autogen/issues/5518

    2025

  20. [20]

    2025.Planning step can make next ActionStep miss its start.https://github.com/h uggingface/smolagents/issues/1097

    2025

  21. [21]

    2025.Runtime context is not being passed to the subgraph.https://github.com/lan gchain-ai/langgraph/issues/5700

    2025

  22. [22]

    2025.Some agents do not deserialize model_context (set to None or Do not serialize too).https://github.com/microsoft/autogen/issues/6336

    2025

  23. [23]

    2025.Structured logging is missing data from Trace logging (0.6.4).https://github .com/microsoft/autogen/issues/6855

    2025

  24. [24]

    2025.Unable to handle inconsistent tool call indices with streaming output when using Qwen3.https://github.com/langchain-ai/langchain/issues/31511

    2025

  25. [25]

    2025.Workflow with multiple cycles stop execute.https://github.com/microsoft/a utogen/issues/6710

    2025

  26. [26]

    2026.AIMessage.tool_calls lost during serialization in InMemoryChatMessageHis- tory.https://github.com/langchain-ai/langchain/issues/34925

    2026

  27. [27]

    https://github.com/deepset-ai/haystack

    2026.Haystack. https://github.com/deepset-ai/haystack

    2026

  28. [28]

    https://pypi.org/project/litellm

    2026.LiteLLM. https://pypi.org/project/litellm

    2026

  29. [29]

    Anonymous. 2026. Understanding Bugs in Modern Agentic Frameworks (Artifact). https://zenodo.org/records/19228789

    work page arXiv 2026

  30. [30]

    Junjie Chen, Yihua Liang, Qingchao Shen, Jiajun Jiang, and Shuochuan Li. 2023. Toward Understanding Deep Learning Framework Bugs.ACM Trans. Softw. Eng. Methodol.32, 6, Article 135 (Sept. 2023), 31 pages. doi:10.1145/3587155

    work page doi:10.1145/3587155 2023

  31. [31]

    2013.Statistical power analysis for the behavioral sciences

    Jacob Cohen. 2013.Statistical power analysis for the behavioral sciences. routledge

    2013

  32. [32]

    Hana Derouiche, Zaki Brahmi, and Haithem Mazeni. 2025. Agentic ai frameworks: Architectures, protocols, and design challenges.arXiv preprint arXiv:2508.10146 (2025)

    work page arXiv 2025

  33. [33]

    Jiawei Han, Jian Pei, Yiwen Yin, and Runying Mao. 2004. Mining frequent patterns without candidate generation: A frequent-pattern tree approach.Data mining and knowledge discovery8, 1 (2004), 53–87

    2004

  34. [34]

    Mohammed Mehedi Hasan, Hao Li, Emad Fallahzadeh, Gopi Krishnan Rajbahadur, Bram Adams, and Ahmed E Hassan. 2025. An empirical study of testing practices in open source AI agent frameworks and agentic applications.arXiv preprint arXiv:2509.19185(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    Niful Islam, Ragib Shahriar Ayon, Deepak George Thomas, Shibbir Ahmed, and Mohammad Wardat. 2026. When Agents Fail: A Comprehensive Study of Bugs in LLM Agents with Automated Labeling.arXiv preprint arXiv:2601.15232(2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  36. [36]

    Zhenmin Li, Lin Tan, Xuanhui Wang, Shan Lu, Yuanyuan Zhou, and Chengxiang Zhai. 2006. Have things changed now? an empirical study of bug characteristics in modern open source software. InProceedings of the 1st Workshop on Architectural and System Support for Improving Software Dependability(San Jose, California) (ASID ’06). Association for Computing Machi...

    work page doi:10.1145/1181309.1181314 2006

  37. [37]

    Siddique, and Umar Farooq

    Daniel Liu, Krishna Upadhyay, Vinaik Chhetri, A.B. Siddique, and Umar Farooq

  38. [38]

    WaveGNN: Integrating graph neural networks and transformers for decay-aware classification of irregular clinical time-series

    A Large-Scale Study on the Development and Issues of Multi-Agent AI Systems. In2025 IEEE International Conference on Big Data (BigData). 7785–7792. doi:10.1109/BigData66926.2025.11402104

    work page doi:10.1109/bigdata66926.2025.11402104 2025

  39. [39]

    2022.LlamaIndex

    Jerry Liu. 2022.LlamaIndex. doi:10.5281/zenodo.1234

    work page doi:10.5281/zenodo.1234 2022

  40. [40]

    Mugeng Liu, Siqi Zhong, Weichen Bi, Yixuan Zhang, Zhiyang Chen, Zhenpeng Chen, Xuanzhe Liu, and Yun Ma. 2026. A First Look at Bugs in LLM Inference Engines.ACM Trans. Softw. Eng. Methodol.(Jan. 2026). doi:10.1145/3788873 Just Accepted

    work page doi:10.1145/3788873 2026

  41. [41]

    Ruofan Lu, Yichen Li, and Yintong Huo. 2025. Exploring Autonomous Agents: A Closer Look at Why They Fail When Completing Tasks. In2025 40th IEEE/ACM International Conference on Automated Software Engineering (ASE). 3856–3860. doi:10.1109/ASE63991.2025.00330

    work page doi:10.1109/ase63991.2025.00330 2025

  42. [42]

    Menéndez, J.A

    M.L. Menéndez, J.A. Pardo, L. Pardo, and M.C. Pardo. 1997. The Jensen-Shannon divergence.Journal of the Franklin Institute334, 2 (1997), 307–318. doi:10.1016/ S0016-0032(96)00063-4

    1997

  43. [43]

    Gonzalez, Matei Zaharia, and Ion Stoica

    Melissa Z Pan, Mert Cemri, Lakshya A Agrawal, Shuyi Yang, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Kannan Ramchandran, Dan Klein, Joseph E. Gonzalez, Matei Zaharia, and Ion Stoica. 2025. Why Do Multiagent Systems Fail?. InICLR 2025 Workshop on Building Trust in Language Models and Applications. https://openreview.net/forum?id=wM521FqPvI

    2025

  44. [44]

    Rui Ren, Jinheng Li, Yan Yin, and Shuai Tian. 2021. Failure Prediction for Large- Scale Clusters Logs via Mining Frequent Patterns. InIntelligent Computing and Block Chain, Wanling Gao, Kai Hwang, Changyun Wang, Weiping Li, Zhigang Qiu, Lei Wang, Aoying Zhou, Weining Qian, Cheqing Jin, and Zhifei Zhang (Eds.). Springer Singapore, Singapore, 147–165

    2021

  45. [45]

    Aymeric Roucher, Albert Villanova del Moral, Thomas Wolf, Leandro von Werra, and Erik Kaunismäki. 2025. Smolagents: a smol library to build great agentic systems. https://github.com/huggingface/smolagents

    2025

  46. [46]

    Guogen Shan and Shawn Gerstenberger. 2017. Fisher’s exact approach for post hoc analysis of a chi-squared test.PloS one12, 12 (2017), e0188709

    2017

  47. [47]

    Qingchao Shen, Haoyang Ma, Junjie Chen, Yongqiang Tian, Shing-Chi Cheung, and Xiang Chen. 2021. A comprehensive study of deep learning compiler bugs. InProceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering(Athens, Greece)(ESEC/FSE 2021). Association for Computing Mach...

    work page doi:10.1145/3468264.3468591 2021

  48. [48]

    In: International Conference on Fuzzy Systems, pp

    Susana M. Vieira, Uzay Kaymak, and João M. C. Sousa. 2010. Cohen’s kappa coef- ficient as a performance measure for feature selection. InInternational Conference on Fuzzy Systems. 1–8. doi:10.1109/FUZZY.2010.5584447

    work page doi:10.1109/fuzzy.2010.5584447 2010

  49. [49]

    Haibo Wang, Zhuolin Xu, Huaien Zhang, Nikolaos Tsantalis, and Shin Hwei Tan

  50. [50]

    Towards Understanding Refactoring Engine Bugs.ACM Trans. Softw. Eng. Methodol.(July 2025). doi:10.1145/3747289 Just Accepted

    work page doi:10.1145/3747289 2025

  51. [51]

    Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, Ahmed Hassan Awadallah, Ryen W White, Doug Burger, and Chi Wang. 2024. AutoGen: Enabling Next- Gen LLM Applications via Multi-Agent Conversations. InFirst Conference on Language Modeling. https://openreview.net/forum?id=BAakY1hNKS

    2024

  52. [52]

    Yiheng Xiong, Mengqian Xu, Ting Su, Jingling Sun, Jue Wang, He Wen, Geguang Pu, Jifeng He, and Zhendong Su. 2023. An Empirical Study of Functional Bugs in Android Apps. InProceedings of the 32nd ACM SIGSOFT International Symposium on Software Testing and Analysis(Seattle, WA, USA)(ISSTA 2023). Association for Computing Machinery, New York, NY, USA, 1319–1...

    work page doi:10.1145/3597926.35 2023

  53. [53]

    Ziluo Xue, Yanjie Zhao, Shenao Wang, Kai Chen, and Haoyu Wang. 2025. A Characterization Study of Bugs in LLM Agent Workflow Orchestration Frame- works. In2025 40th IEEE/ACM International Conference on Automated Software Engineering (ASE). 3369–3380. doi:10.1109/ASE63991.2025.00278

    work page doi:10.1109/ase63991.2025.00278 2025

  54. [54]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. InInternational Conference on Learning Representations (ICLR)

    2023

  55. [55]

    Daojuan Zhang, Tianqi Wu, Xiaoming Zhou, Bo Hu, and Wenjie Zhang. 2024. Multi-source System Log Behavior Pattern Mining Method Based on FP-Growth. InProceedings of the 2023 International Conference on Communication Network and Machine Learning(Zhengzhou, China)(CNML ’23). Association for Computing Machinery, New York, NY, USA, 248–254. doi:10.1145/3640912.3640961

    work page doi:10.1145/3640912.3640961 2024

  56. [56]

    Huaien Zhang, Yu Pei, Shuyun Liang, and Shin Hwei Tan. 2024. Understanding and Detecting Annotation-Induced Faults of Static Analyzers.Proc. ACM Softw. Eng.1, FSE, Article 33 (July 2024), 23 pages. doi:10.1145/3643759

    work page doi:10.1145/3643759 2024