Elamur Research⌕
☰
home / papers / 2026 / Metacognitive Self-Correction for Multi-Agent System via Prototype-Guided Next-Execution Reconstruction

Metacognitive Self-Correction for Multi-Agent System via Prototype-Guided Next-Execution Reconstruction

Paper · 2026 · Xu Shen, Qi Zhang, Song Wang et al.

Summary

LLM-based multi-agent systems (MAS) are fragile: one agent's mistake can spread through the collaboration and derail the whole run. In the authors' fault-injection tests, this lowered accuracy by as much as 51.9 points on AQuA. Step-level error labels are expensive, and isolated steps are hard to classify without context. The authors therefore propose MASC, which detects and fixes errors online without supervision. Detection is treated as anomaly scoring conditioned on the interaction history. A Next-Execution Reconstruction module predicts the embedding of the upcoming step from the query and prior turns, so a mismatch signals a likely error. A Prototype-Guided Enhancement learns a reference representation of normal steps, which keeps scores stable early in a trajectory when little history exists. When a step is flagged, a correction agent rewrites that output before later agents receive it. MASC beats supervised and unsupervised baselines on Who&When and AgentErrorBench. As a plug-in, it adds modest but consistent accuracy across several MAS topologies. This matters because it improves reliability without costly annotation or retraining.

Key points

  • On Who&When in the "w/o GT" setting, MASC reaches 77.84% AUC-ROC on the handcrafted subset and 75.62% on the algorithm-generated subset, up to 8.47% above the best baseline.
  • On AgentErrorBench, MASC gains up to 2.5% absolute on GAIA trajectories and 6.0% on WebShop over the previous state of the art.
  • Added to Chain, Complete, Random and LLM-Debate frameworks, MASC improves all six benchmarks (MMLU, GSM8K, AQuA, MultiArith, SVAMP, HumanEval), by 1.29% on average. LLM-Debate, for example, rises from 87.53% to 88.89%.
  • On GSM8K correction, MASC adds 2.46 points on average (up to 3.15). The Step-by-Step, BERT classifier and LLM classifier detectors give unstable gains or even hurt accuracy.
  • Ablations show both modules are needed: removing reconstruction or the prototype lowers AUC-ROC, and the prototype matters most for early steps.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of North Carolina at Chapel HillChapel Hill · United States67 works · 16,474 citations🎓Arizona State UniversityTempe · United States50 works · 2,032 citations🎓Michigan State UniversityEast Lansing · United States23 works · 1,077 citations🎓University of Central FloridaOrlando · United States13 works · 39,673 citations🎓Temple UniversityPhiladelphia · United States10 works · 24,319 citations

Also on the byline: Honda (United States) · Honda Research Institute, USA.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

Cite

Cite as Xu Shen, Qi Zhang, Song Wang, Zhen Tan, Xinyu Zhao, Laura Yao, Vaishnav Tadiparthi, Hossein Nourkhiz Mahjoub, Ehsan Moradi Pari, Kwonjoon Lee and Tianlong Chen. 2026. Metacognitive Self-Correction for Multi-Agent System via Prototype-Guided Next-Execution Reconstruction. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1168

@inproceedings{shen2026metacognitive,
  title = {{Metacognitive Self-Correction for Multi-Agent System via Prototype-Guided Next-Execution Reconstruction}},
  author = {Xu Shen, Qi Zhang, Song Wang, Zhen Tan, Xinyu Zhao, Laura Yao, Vaishnav Tadiparthi, Hossein Nourkhiz Mahjoub, Ehsan Moradi Pari, Kwonjoon Lee and Tianlong Chen},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1168},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1168}
}

References

65 references, 33 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Maksym Andriushchenko, Alexandra Souly, Mateusz Dziemian, Derek Duenas, Maxwell Lin, Justin Wang, Dan Hendrycks, Andy Zou, J Zico Kolter, Matt Fredrikson, and 1 others. 2025. Agentharm: A benchmark for measuring harmfulness of llm agents. In The Thirteenth International Conference on Learning Repres… find
  2. Parishad BehnamGhader, Vaibhav Adlakha, Marius Mosbach, Dzmitry Bahdanau, Nicolas Chapados, and Siva Reddy. 2024. Llm2vec: Large language models are secretly powerful text encoders. arXiv preprint arXiv:2404.05961.
  3. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  4. Mert Cemri, Melissa Z Pan, Shuyi Yang, Lakshya A Agrawal, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Dan Klein, Kannan Ramchandran, and 1 others. 2025. Why do multi-agent llm systems fail? arXiv preprint arXiv:2503.13657. find
  5. Jun Shern Chan, Neil Chowdhury, Oliver Jaffe, James Aung, Dane Sherburn, Evan Mays, Giulio Starace, Kevin Liu, Leon Maksin, Tejal Patwardhan, and 1 others. Mle-bench: Evaluating machine learning agents on machine learning engineering. In The Thirteenth International Conference on Learning Representa…
  6. Lingjiao Chen, Jared Quincy Davis, Boris Hanin, Peter Bailis, Matei Zaharia, James Zou, and Ion Stoica. 2025. Optimizing model selection for compound ai systems. arXiv preprint arXiv:2502.14815. find
  7. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, and 39 others. 202… — Evaluating Large Language Models Trained on Code
  8. Pingjun Hong, Beiduo Chen, Siyao Peng, Marie-Catherine de Marneffe, Benjamin Roth, and Barbara Plank. 2025b. Agree, disagree, explain: Decomposing human label variation in nli through the lens of explanations. arXiv preprint arXiv:2510.16458.
  9. Zhaorun Chen, Zhen Xiang, Chaowei Xiao, Dawn Song, and Bo Li. 2024. Agentpoison: Red-teaming llm agents via poisoning memory or knowledge bases. Advances in Neural Information Processing Systems, 37:130185–130213. find
  10. Pingjun Hong and Benjamin Roth. 2026. Do llm self-explanations help users predict model behavior? evaluating counterfactual simulatability with pragmatic perturbations. arXiv preprint arXiv:2601.03775. — [arXiv:2601.03775](https://arxiv.org/abs/2601.03775) find
  11. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. Training verifiers to solve math word problems. arXiv prepring, abs/2110.14168.
  12. Jen-tse Huang, Eric John Li, Man Ho Lam, Tian Liang, Wenxuan Wang, Youliang Yuan, Wenxiang Jiao, Xing Wang, Zhaopeng Tu, and Michael R. Lyu. 2025. Competing large language models in multi-agent gaming environments. In Proceedings of the Thirteenth International Conference on Learning Representations… find
  13. Yufan Dang, Chen Qian, Xueheng Luo, Jingru Fan, Zihao Xie, Ruijie Shi, Weize Chen, Cheng Yang, Xiaoyin Che, Ye Tian, and 1 others. 2025. Multi-agent collaboration via evolving orchestration. arXiv preprint arXiv:2505.19591. — [arXiv:2505.19591](https://arxiv.org/abs/2505.19591) find
  14. Yoichi Ishibashi and Yoshimasa Nishimura. 2024. Self-organized agents: A llm multi-agent framework toward ultra large-scale code generation and optimization. arXiv preprint arXiv:2404.02183. — [arXiv:2404.02183](https://arxiv.org/abs/2404.02183) find
  15. Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, and Igor Mordatch. 2023. Improving factuality and reasoning in language models through multiagent debate. CoRR, abs/2305.14325.
  16. Yuyou Gan, Yong Yang, Zhe Ma, Ping He, Rui Zeng, Yiming Wang, Qingming Li, Chunyi Zhou, Songze Li, Ting Wang, and 1 others. 2024. Navigating the risks: A survey of security, privacy, and ethics threats in llm-based agents. arXiv preprint arXiv:2411.09523. — [arXiv:2411.09523](https://arxiv.org/abs/2411.09523) find
  17. Fanqi Kong, Ruijie Zhang, Huaxiao Yin, Guibin Zhang, Xiaofei Zhang, Ziang Chen, Zhaowei Zhang, Xiaoyuan Zhang, Song-Chun Zhu, and Xue Feng. 2025. Aegis: Automated error generation and attribution for multi-agent systems. arXiv preprint arXiv:2509.14295. — [arXiv:2509.14295](https://arxiv.org/abs/2509.14295) find
  18. Mikhail V Koroteev. 2021. Bert: a review of applications in natural language processing and understanding. arXiv preprint arXiv:2103.11943. find
  19. Alireza Ghafarollahi and Markus J Buehler. 2024. Sciagents: Automating scientific discovery through multi-agent intelligent graph reasoning. arXiv preprint arXiv:2409.05556. find
  20. Zhenyu Lei, Zhen Tan, Song Wang, Yaochen Zhu, Zihan Chen, Yushun Dong, and Jundong Li. 2025. Learning from diverse reasoning paths with routing and collaboration. arXiv preprint arXiv:2508.16861.
  21. Junda He, Christoph Treude, and David Lo. 2025a. Llm-based multi-agent systems for software engineering: Literature review, vision, and the road ahead. ACM Transactions on Software Engineering and Methodology, 34(5):1–30. find
  22. Dawei Li, Zhen Tan, Peijia Qian, Yifan Li, Kumar Chaudhary, Lijie Hu, and Jiayi Shen. 2025a. Smoa: Improving multi-agent large language models with s parse m ixture-o f-a gents. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pages 54–65. Springer. find
  23. Pengfei He, Zhenwei Dai, Xianfeng Tang, Yue Xing, Hui Liu, Jingying Zeng, Qiankun Peng, Shrivats Agrawal, Samarth Varshney, Suhang Wang, and 1 others. 2025b. Attention knows whom to trust: Attention-based trust management for llm multi-agent systems. arXiv preprint arXiv:2506.02546. — [arXiv:2506.02546](https://arxiv.org/abs/2506.02546) find
  24. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. Measuring massive multitask language understanding. Proceedings of the International Conference on Learning Representations (ICLR).
  25. Pingjun Hong, Beiduo Chen, Siyao Peng, Marie-Catherine de Marneffe, and Barbara Plank. 2025a. Litex: A linguistic taxonomy of explanations for understanding within-label variation in natural language inference. arXiv preprint arXiv:2505.22848.
  26. Guohao Li, Hasan Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. 2023. Camel: Communicative agents for" mind" exploration of large language model society. Advances in Neural Information Processing Systems, 36:51991–52008.
  27. Shiyuan Li, Yixin Liu, Qingsong Wen, Chengqi Zhang, and Shirui Pan. 2025b. Assemble your crew: Automatic multi-agent communication topology design via autoregressive graph generation. arXiv preprint arXiv:2507.18224. — [arXiv:2507.18224](https://arxiv.org/abs/2507.18224) find
  28. Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. 2017. Program induction by rationale generation: Learning to solve and explain algebraic word problems. arXiv preprint arXiv:1705.04146. find
  29. Rui Miao, Yixin Liu, Yili Wang, Xu Shen, Yue Tan, Yiwei Dai, Shirui Pan, and Xin Wang. 2025. Blindguard: Safeguarding llm-based multi-agent systems under unknown attacks. arXiv preprint arXiv:2508.08127.
  30. Song Wang, Zihan Chen, Peng Wang, Zhepei Wei, Zhen Tan, Yu Meng, Cong Shen, and Jundong Li. 2025b. Separate the wheat from the chaff: Winnowing down divergent views in retrieval augmented generation. In EMNLP 2025. find
  31. Fan Nie, Lan Feng, Haotian Ye, Weixin Liang, Pan Lu, Huaxiu Yao, Alexandre Alahi, and James Zou. 2025. Weak-for-strong: Training weak meta-agent to harness strong executors. arXiv preprint arXiv:2504.04785. find
  32. Song Wang, Zhen Tan, Zihan Chen, Shuang Zhou, Tianlong Chen, and Jundong Li. 2025c. Anymac: Cascading flexible multi-agent collaboration via next-agent prediction. In EMNLP 2025.
  33. Arkil Patel, Satwik Bhattamishra, and Navin Goyal. 2021. Are nlp models really able to solve simple math word problems? arXiv preprint arXiv:2103.07191.
  34. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171.
  35. Chen Qian, Wei Liu, Hongzhang Liu, Nuo Chen, Yufan Dang, Jiahao Li, Cheng Yang, Weize Chen, Yusheng Su, Xin Cong, and 1 others. 2023. Chatdev: Communicative agents for software development. arXiv preprint arXiv:2307.07924.
  36. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems, pages 24824–24837.
  37. Chen Qian, Zihao Xie, Yifei Wang, Wei Liu, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, Zhiyuan Liu, and Maosong Sun. 2024. Scaling large-language-model-based multi-agent collaboration. arXiv preprint arXiv:2406.07155. find
  38. Jinyang Wu, Chonghua Liao, Mingkuan Feng, Shuai Zhang, Zhengqi Wen, Haoran Luo, Ling Yang, Huazhe Xu, and Jianhua Tao. 2025. Templaterl: Structured template-guided reinforcement learning for llm reasoning. arXiv preprint arXiv:2505.15692.
  39. Subhro Roy and Dan Roth. 2016. Solving general arithmetic word problems. arXiv preprint arXiv:1608.01413.
  40. Samuel Schmidgall, Yusheng Su, Ze Wang, Ximeng Sun, Jialian Wu, Xiaodong Yu, Jiang Liu, Michael Moor, Zicheng Liu, and Emad Barsoum. 2025. Agent laboratory: Using llm agents as research assistants. arXiv preprint arXiv:2501.04227.
  41. Xu Shen, Yixin Liu, Yiwei Dai, Yili Wang, Rui Miao, Yue Tan, Shirui Pan, and Xin Wang. 2025a. Understanding the information propagation effects of communication topologies in llm-based multi-agent systems. arXiv preprint arXiv:2505.23352.
  42. Xu Shen, Song Wang, Zhen Tan, Laura Yao, Xinyu Zhao, Kaidi Xu, Xin Wang, and Tianlong Chen. 2025b. Faithcot-bench: Benchmarking instance-level faithfulness of chain-of-thought reasoning. arXiv preprint arXiv:2510.04040. — [arXiv:2510.04040](https://arxiv.org/abs/2510.04040) find
  43. Zhen Tan, Jun Yan, I Hsu, Rujun Han, Zifeng Wang, Long T Le, Yiwen Song, Yanfei Chen, Hamid Palangi, George Lee, and 1 others. 2025. In prospect and retrospect: Reflective memory management for long-term personalized dialogue agents. arXiv preprint arXiv:2503.08026.
  44. Shilong Wang, Guibin Zhang, Miao Yu, Guancheng Wan, Fanci Meng, Chongye Guo, Kun Wang, and Yang Wang. 2025a. G-safeguard: A topology-guided security lens and treatment on llm-based multi-agent systems. arXiv preprint arXiv:2502.11127.
  45. Jinyang Wu, Shuo Yang, Changpeng Yang, Yuhao Shen, Shuai Zhang, Zhengqi Wen, and Jianhua Tao. 2026a. Spark: Strategic policy-aware exploration via dynamic branching for long-horizon agentic learning. arXiv preprint arXiv:2601.20209.
  46. Jinyang Wu, Guocheng Zhai, Ruihan Jin, Jiahao Yuan, Yuhao Shen, Shuai Zhang, Zhengqi Wen, and Jianhua Tao. 2026b. Atlas: Orchestrating heterogeneous models and tools for multi-domain complex reasoning. arXiv preprint arXiv:2601.03872.
  47. Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Shaokun Zhang, Erkang Zhu, Beibin Li, Li Jiang, Xiaoyun Zhang, and Chi Wang. 2023. Autogen: Enabling next-gen llm applications via multi-agent conversation framework. Conference on Language Modeling.
  48. Chengxing Xie, Canyu Chen, Feiran Jia, Ziyu Ye, Shiyang Lai, Kai Shu, Jindong Gu, Adel Bibi, Ziniu Hu, David Jurgens, and 1 others. 2024. Can large language model agents simulate human trust behavior? Advances in neural information processing systems, 37:15674–15729. find
  49. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023a. Tree of thoughts: Deliberate problem solving with large language models. find
  50. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2023b. React: Synergizing reasoning and acting in language models. In The Eleventh International Conference on Learning Representations.
  51. Zihao Yi, Jiarui Ouyang, Yuwen Liu, Tianhao Liao, Zhe Xu, and Ying Shen. 2024. A survey on recent advances in llm-based multi-turn dialogue systems. arXiv preprint arXiv:2402.18013.
  52. Zhuosheng Zhang, Aston Zhang, Mu Li, and Alex Smola. 2022. Automatic chain of thought prompting in large language models. arXiv preprint arXiv:2210.03493.
  53. Miao Yu, Shilong Wang, Guibin Zhang, Junyuan Mao, Chenlong Yin, Qijiong Liu, Qingsong Wen, Kun Wang, and Yang Wang. 2024. Netsafe: Exploring the topological safety of multi-agent networks. arXiv preprint arXiv:2410.15686. — [arXiv:2410.15686](https://arxiv.org/abs/2410.15686) find
  54. Kunlun Zhu, Zijia Liu, Bingxuan Li, Muxin Tian, Yingxuan Yang, Jiaxun Zhang, Pengrui Han, Qipeng Xie, Fuyang Cui, Weijia Zhang, and 1 others. 2025. Where llm agents fail and how they can learn from failures. arXiv preprint arXiv:2509.25370. — [arXiv:2509.25370](https://arxiv.org/abs/2509.25370) find
  55. Tongxin Yuan, Zhiwei He, Lingzhong Dong, Yiming Wang, Ruijie Zhao, Tian Xia, Lizhen Xu, Binglin Zhou, Fangqi Li, Zhuosheng Zhang, and 1 others. 2024. R-judge: Benchmarking safety risk awareness for llm agents. arXiv preprint arXiv:2401.10019.
  56. Longfei Zuo, Pingjun Hong, Oliver Kraus, Barbara Plank, and Robert Litschko. 2025a. Evaluating large language models for cross-lingual retrieval. arXiv preprint arXiv:2509.14749. find
  57. Qiusi Zhan, Zhixiang Liang, Zifan Ying, and Daniel Kang. 2024. Injecagent: Benchmarking indirect prompt injections in tool-integrated large language model agents. In Findings of the Association for Computational Linguistics ACL 2024, pages 10471– 10506.
  58. Longfei Zuo, Barbara Plank, and Siyao Peng. 2025b. Evade: Llm-based explanation generation and validation for error detection in nli. arXiv preprint arXiv:2511.08949.
  59. Guibin Zhang, Kaijie Chen, Guancheng Wan, Heng Chang, Hong Cheng, Kun Wang, Shuyue Hu, and Lei Bai. 2025a. Evoflow: Evolving diverse agentic workflows on the fly. arXiv preprint arXiv:2502.07373. — [arXiv:2502.07373](https://arxiv.org/abs/2502.07373) find
  60. Guibin Zhang, Luyang Niu, Junfeng Fang, Kun Wang, Lei Bai, and Xiang Wang. 2025b. Multi-agent architecture search via agentic supernet. arXiv preprint arXiv:2502.04180. — [arXiv:2502.04180](https://arxiv.org/abs/2502.04180) find
  61. Guibin Zhang, Junhao Wang, Junjie Chen, Wangchunshu Zhou, Kun Wang, and Shuicheng Yan. 2025c. Agentracer: Who is inducing failure in the llm agentic systems? arXiv preprint arXiv:2509.03312. — [arXiv:2509.03312](https://arxiv.org/abs/2509.03312) find
  62. Guibin Zhang, Yanwei Yue, Xiangguo Sun, Guancheng Wan, Miao Yu, Junfeng Fang, Kun Wang, Tianlong Chen, and Dawei Cheng. 2024a. G-designer: Architecting multi-agent communication topologies via graph neural networks. arXiv preprint arXiv:2410.11782. — [arXiv:2410.11782](https://arxiv.org/abs/2410.11782) find
  63. Kechi Zhang, Jia Li, Ge Li, Xianjie Shi, and Zhi Jin. 2024b. CodeAgent: Enhancing code generation with tool-integrated agent systems for real-world repo-level coding challenges. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages…
  64. Shaokun Zhang, Ming Yin, Jieyu Zhang, Jiale Liu, Zhiguang Han, Jingyang Zhang, Beibin Li, Chi Wang, Huazheng Wang, Yiran Chen, and 1 others. 2025d. Which agent causes task failures and when? on automated failure attribution of llm multi-agent systems. In Forty-second International Conference on Mach… find
  65. Wentao Zhang, Liang Zeng, Yuzhen Xiao, Yongcong Li, Ce Cui, Yilei Zhao, Rui Hu, Yang Liu, Yahui Zhou, and Bo An. 2025e. Agentorchestra: A hierarchical multi-agent framework for general-purpose task solving. arXiv preprint arXiv:2506.12508. — [arXiv:2506.12508](https://arxiv.org/abs/2506.12508) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shen-2026-metacognitive-self-correction-multi-agent/