pith. sign in

arxiv: 2607.00041 · v1 · pith:HDOXYQ34new · submitted 2026-06-29 · 💻 cs.SE · cs.AI

ATM: CID-Brokered Pre-Write Admission for Multi-Agent Code Co-Synthesis

Eagl Huang This is my paper

Pith reviewed 2026-07-02 20:34 UTC · model grok-4.3

classification 💻 cs.SE cs.AI
keywords multi-agent systemscode co-synthesiswrite admissionCID brokergovernance chainsemantic atomsvirtual atomsshared mutation
0
0 comments X

The pith

ATM binds task intent, scope, admission, validation and evidence into one chain using a CID broker for shared-mutation decisions.

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

Multi-agent LLM systems decompose software work into planning, generation and repair but still need a mechanism to decide which concurrent write intents may run in parallel, which must serialize, and which must fail closed. ATM creates a single governance chain that links task intent to repository scope, write admission, validation and evidence obligations. A CID broker performs the admission step by mapping intents to semantic atoms through adapter-guided atomization; virtual atoms supply temporary units when persistent coverage is incomplete. Writes are executed by a neutral steward rather than the proposing agents. Controlled scenarios, archived cases, an admission benchmark, a three-week adopter study and a recovery benchmark supply evidence of feasibility, auditability and bounded recoverability inside observed single-domain settings.

Core claim

The AI-Atomic-Framework (ATM) binds task intent, repository scope, write admission, validation, and evidence obligations into one governance chain. A Content Identifier (CID) broker serves as the shared-mutation admission subsystem. Adapter-guided atomization maps write intents to semantic atoms and bounded regions; when persistent atom-map coverage is incomplete, virtual atoms provide temporary auditable governance units for conservative comparison and routing. Governed shared writes are ultimately applied by a neutral steward rather than directly by proposing agents.

What carries the argument

The CID broker as shared-mutation admission subsystem that routes intents via adapter-guided atomization and virtual atoms.

If this is right

  • Write intents receive deterministic routing to parallel, serialized or fail-closed paths before any mutation occurs.
  • A neutral steward, not the proposing agents, applies all governed writes.
  • Virtual atoms maintain auditability when persistent atom-map coverage is incomplete.
  • The same governance chain supports both feasibility checks and bounded recovery in single-domain operation.
  • Evidence from a 12-scenario design matrix, ATM-AdmissionBench and external-adopter study is consistent with these properties.

Where Pith is reading between the lines

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

  • The single-governance-domain restriction may limit direct use in federated or cross-repository agent teams.
  • Virtual atoms could be tested as a general pattern for any collaborative editing system that lacks complete static maps.
  • Integration of the CID broker with existing version-control hooks would allow empirical measurement of conflict reduction in live multi-agent workflows.

Load-bearing premise

Adapter-guided atomization plus virtual atoms can supply conservative, auditable routing when persistent atom-map coverage is incomplete without introducing new failure modes or requiring cross-domain coordination.

What would settle it

An observed case in which an incomplete atom map produces an unsafe parallel admission or an unrecoverable state inside a single governance domain.

read the original abstract

Multi-agent LLM systems can decompose software-engineering work into planning, generation, validation, and repair, but a narrower systems problem remains: before any governed shared mutation is applied, a system must decide which concurrently formed write intents may proceed in parallel, which require deterministic composition or serialization, and which must take a fail-closed path. We address this problem with the AI-Atomic-Framework (ATM), a specification-grounded governance substrate for software agents operating within a single governance domain. ATM binds task intent, repository scope, write admission, validation, and evidence obligations into one governance chain. A Content Identifier (CID) broker serves as the shared-mutation admission subsystem. Adapter-guided atomization maps write intents to semantic atoms and bounded regions; when persistent atom-map coverage is incomplete, virtual atoms provide temporary auditable governance units for conservative comparison and routing. Governed shared writes are ultimately applied by a neutral steward rather than directly by proposing agents. Evaluation combines controlled, field, adoption, and extension evidence, including a 12-scenario deterministic design matrix, three archived runner cases, ATM-AdmissionBench, three archived same-file boundary cases, a three-week external-adopter study, and an operational recovery-routing benchmark. The results support feasibility, auditability, and bounded recoverability within the observed single-domain settings, but do not claim broad comparative superiority or cross-clone governance.

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

2 major / 1 minor

Summary. The paper introduces the AI-Atomic-Framework (ATM) as a specification-grounded governance substrate for multi-agent LLM systems performing software-engineering tasks. It binds task intent, repository scope, write admission, validation, and evidence obligations into a single chain, with a CID broker acting as the shared-mutation admission subsystem. Adapter-guided atomization maps intents to semantic atoms and bounded regions; virtual atoms handle incomplete persistent atom-map coverage via temporary auditable units for conservative comparison and routing. Governed writes are applied by a neutral steward. Evaluation draws on a 12-scenario deterministic design matrix, three archived runner cases, ATM-AdmissionBench, three archived same-file boundary cases, a three-week external-adopter study, and an operational recovery-routing benchmark; the results are presented as supporting feasibility, auditability, and bounded recoverability within observed single-domain settings, without claims of broad superiority or cross-clone applicability.

Significance. If the central claims hold, the work supplies a concrete governance mechanism for concurrent write management in multi-agent code synthesis, addressing a practical systems gap between decomposition and safe shared mutation. The multi-method evaluation approach (controlled matrix, archived cases, external study, and benchmark) is a positive feature that strengthens the feasibility argument within the stated single-domain scope.

major comments (2)
  1. [Abstract] Abstract (paragraph on virtual atoms): the claim that virtual atoms deliver 'conservative comparison and routing' when persistent atom-map coverage is incomplete is load-bearing for the CID broker's admission subsystem and the overall governance chain, yet the manuscript provides no algorithmic specification of atom construction, comparison logic, or decision rules. Without this, it is impossible to verify absence of over-admission (false negatives on conflicts) or under-admission that would break the stated bounded recoverability.
  2. [Abstract] Abstract (evaluation paragraph): the central claim that the listed evidence 'supports feasibility, auditability, and bounded recoverability' rests on unshown quantitative results, error bars, or derivation details from the 12-scenario design matrix, ATM-AdmissionBench, or recovery-routing benchmark. This absence prevents assessment of whether the observed single-domain results actually substantiate the recoverability guarantee.
minor comments (1)
  1. [Abstract] The abstract introduces several new entities (CID broker, semantic atoms, virtual atoms) without a brief forward reference to their definitions or invariants in the main text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and the major revision recommendation. We address each comment below. Both points correctly identify missing details in the current manuscript, and we will revise to supply them.

read point-by-point responses

  1. Referee: [Abstract] Abstract (paragraph on virtual atoms): the claim that virtual atoms deliver 'conservative comparison and routing' when persistent atom-map coverage is incomplete is load-bearing for the CID broker's admission subsystem and the overall governance chain, yet the manuscript provides no algorithmic specification of atom construction, comparison logic, or decision rules. Without this, it is impossible to verify absence of over-admission (false negatives on conflicts) or under-admission that would break the stated bounded recoverability.

    Authors: The referee correctly notes the absence of algorithmic specification for virtual atom construction, comparison logic, and decision rules. The current manuscript does not contain these details. In revision we will add a dedicated subsection with pseudocode, formal rules, and decision procedures to enable verification of conservative properties and bounded recoverability. revision: yes

  2. Referee: [Abstract] Abstract (evaluation paragraph): the central claim that the listed evidence 'supports feasibility, auditability, and bounded recoverability' rests on unshown quantitative results, error bars, or derivation details from the 12-scenario design matrix, ATM-AdmissionBench, or recovery-routing benchmark. This absence prevents assessment of whether the observed single-domain results actually substantiate the recoverability guarantee.

    Authors: The referee is correct that the abstract and evaluation summary lack the quantitative results, error bars, and derivation details. The manuscript presents the evaluation approach but does not include these specifics in the provided sections. We will revise the abstract and main text to incorporate key quantitative findings, tables, and derivations from the design matrix and benchmarks. revision: yes

Circularity Check

0 steps flagged

No significant circularity in specification-grounded framework

full rationale

The paper presents ATM as a specification-grounded governance substrate that binds task intent, repository scope, write admission, validation, and evidence obligations into one chain, with CID broker and virtual atoms as design elements for conservative routing when atom-map coverage is incomplete. No equations, fitted parameters, predictions, or derivation steps appear in the abstract or described structure. Claims rest on the construction itself plus listed evaluation evidence (design matrix, benchmarks, adopter study) rather than reducing to self-inputs by construction. No self-citation load-bearing uniqueness theorems or ansatz smuggling are referenced. This is a systems design specification, not a predictive derivation chain, so the central claims remain independent of the input patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

The framework rests on the existence of a single governance domain and the ability to map arbitrary write intents to bounded semantic atoms or virtual atoms; these are introduced without external benchmarks or shipped artifacts.

axioms (1)
  • domain assumption A single governance domain exists in which all agents operate and a neutral steward can apply writes.
    Stated in abstract as the setting for all evaluation evidence.
invented entities (3)
  • CID broker no independent evidence
    purpose: Shared-mutation admission subsystem that routes write intents.
    Core new component introduced to solve the admission problem; no independent evidence supplied.
  • semantic atoms no independent evidence
    purpose: Bounded regions for conservative comparison and routing of writes.
    Invented mapping unit; no external validation mentioned.
  • virtual atoms no independent evidence
    purpose: Temporary auditable governance units when atom-map coverage is incomplete.
    Fallback mechanism introduced to handle incomplete mappings; no falsifiable handle outside the paper.

pith-pipeline@v0.9.1-grok · 5772 in / 1339 out tokens · 23793 ms · 2026-07-02T20:34:30.902350+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

44 extracted references · 39 canonical work pages · 16 internal anchors

  1. [1]

    Pugachev, Sergey. 2025. ”CodeCRDT: Observation-Driven Coordination for Multi-Agent LLM Code Generation. ” arXiv:2510.18893 [cs.DC]. https://doi.org/10.48550/arXiv.2510.18893

    work page doi:10.48550/arxiv.2510.18893 2025

  2. [2]

    Acharya, Vivek. 2026. ”Semantic Consensus: Process-Aware Conflict Detection and Resolution for Enterprise Multi-Agent LLM Systems. ” arXiv:2604.16339 [cs.AI].https://doi.org/10.48550/arXiv.2604.16339

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.16339 2026

  3. [3]

    Liu, Mengyang, Taozhi Chen, Zhenhua Xu, Xue Jiang, and Yihong Dong. 2026. ”Multi-agent Collaboration with State Management. ” arXiv:2605.20563 [cs.MA]. https://doi.org/10.48550/arXiv.2605.20563. 38

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.20563 2026

  4. [4]

    Qian, Kaiyang, Xinmin Fang, and Zhengxiong Li. 2026. ”MPAC: A Multi-Principal Agent Coordination Protocol for Interoperable Multi-Agent Collaboration. ” arXiv:2604.09744 [cs.MA].https://doi.org/10.48550/arXiv.2604.09744

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.09744 2026

  5. [5]

    Zhou, Weixing, Zhiyou Wang, Zeshun Peng, Hetian Chen, Yanfeng Zhang, and Ge Yu. 2026. ”ATCC: Adaptive Concurrency Control for Unforeseen Agentic Transactions. ” arXiv:2603.13906 [cs.DB]. https://doi.org/10.48550/arXiv.2603.13906

    work page doi:10.48550/arxiv.2603.13906 2026

  6. [6]

    Agrawal, Shuyi Yang, Bhavya Chopra, Rishabh Tiwari, Kurt Keutzer, Aditya Parameswaran, Kannan Ramchandran, Dan Klein, Joseph E

    Pan, Melissa Z., 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?” In *ICLR 2025 Workshop on Building Trust in Language Models and Applications*. https://openreview.net/forum?...

    2025

  7. [7]

    Sartori, Camilo Chacon. 2026. ”The Specification Gap: Coordination Failure Under Partial Knowledge in Code Agents. ” arXiv:2603.24284 [cs.SE]. https://doi.org/10.48550/arXiv.2603.24284

    work page doi:10.48550/arxiv.2603.24284 2026

  8. [8]

    Ellis, Clarence A., and Simon J. Gibbs. 1989. ”Concurrency Control in Groupware Systems. ” In *Proceedings of the 1989 ACM SIGMOD International Conference on Management of Data*, 399–407. New York: ACM Press. https://doi.org/10.1145/ 67544.66963

    work page arXiv 1989

  9. [9]

    Shapiro, Marc, Nuno Preguica, Carlos Baquero, and Marek Zawirski. 2011. ”Conflict-Free Replicated Data Types. ” In *Stabilization, Safety, and Security of Distributed Systems: 13th International Symposium, SSS 2011*, Lecture Notes in Computer Science 6976, 386-400. Berlin: Springer. https://doi.org/10.1007/978-3-642-24550-3_29

    work page doi:10.1007/978-3-642-24550-3_29 2011

  10. [10]

    T., and John T

    Kung, H. T., and John T. Robinson. 1981. ”On Optimistic Methods for Concurrency Control. ” *ACM Transactions on Database Systems* 6 (2): 213-226. https://doi.org/10.1145/319566.319567

    work page doi:10.1145/319566.319567 1981

  11. [11]

    Lyu, Hongtao, Dingyan Zhang, Mingyu Wu, Xingda Wei, and Haibo Chen. 2026. ”CoAgent: Concurrency Control for Multi-Agent Systems. ” arXiv:2606.15376 [cs.DC].https://doi.org/10.48550/arXiv.2606.15376

    work page doi:10.48550/arxiv.2606.15376 2026

  12. [12]

    Geng, Jiayi, and Graham Neubig. 2026. ”Effective Strategies for Asynchronous Software Engineering Agents. ” arXiv:2603.21489 [cs.CL]. https://doi.org/10.48550/arXiv.2603.21489

    work page doi:10.48550/arxiv.2603.21489 2026

  13. [13]

    Zhang, Qingyu, Junzhe Li, Jiayi Lin, Changhua Luo, and Chenxiong Qian. 2026. ”Rover: Context-aware Conflict Resolution with LLM. ” arXiv:2605.17279 [cs.SE].https://doi.org/10.48550/arXiv.2605.17279

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.17279 2026

  14. [14]

    Ogenrwot, Daniel, and John Businge. 2026. ”AgenticFlict: A Large-Scale Dataset of Merge Conflicts in AI Coding Agent Pull Requests on GitHub. ” arXiv:2604.03551 [cs.SE]. https://doi.org/10.48550/arXiv.2604.03551

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.03551 2026

  15. [15]

    Wang, Yifei, Ruiyin Li, Peng Liang, Qiong Feng, Zengyang Li, Mojtaba Shahin, and Arif Ali Khan. 2026. ”CodeTeam: An LLM- Powered Multi-Agent Framework for Repository-Level Code Generation. ” arXiv:2606.22082 [cs.SE]. https://doi.org/10.485 50/arXiv.2606.22082

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  16. [16]

    Khan, Sajjad. 2026. ”S-Bus: Automatic Read-Set Reconstruction for Multi-Agent LLM State Coordination. ” arXiv:2605.17076 [cs.LG]. https://doi.org/10.48550/arXiv.2605.17076

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.17076 2026

  17. [17]

    Huang, Beichen, Ran Cheng, and Kay Chen Tan. 2025. ”EvoGit: Decentralized Code Evolution via Git-Based Multi-Agent Collabo- ration. ” arXiv:2506.02049 [cs.SE].https://doi.org/10.48550/arXiv.2506.02049

    work page doi:10.48550/arxiv.2506.02049 2025

  18. [18]

    Li, Yang, Siqi Ping, Xiyu Chen, Xiaojian Qi, Zigan Wang, Ye Luo, and Xiaowei Zhang. 2025. ”AgentGit: A Version Control Framework for Reliable and Scalable LLM-Powered Multi-Agent Systems. ” arXiv:2511.00628 [cs.SE]. https://doi.org/10.48550/arXiv .2511.00628

    work page internal anchor Pith review doi:10.48550/arxiv 2025

  19. [20]

    AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation

    Wu, Qingyun, 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. 2023. ”AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation. ” arXiv:2308.08155 [cs.AI].https://doi.org/10.48550/arXiv.2308.08155

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2308.08155 2023

  20. [21]

    Adya, Atul. 1999. ”Weak Consistency: A Generalized Theory and Optimistic Implementations for Distributed Transactions. ” PhD thesis, Massachusetts Institute of Technology. https://hdl.handle.net/1721.1/149899

    1999

  21. [22]

    Freedman, Michael Kaminsky, and David G

    Lloyd, Wyatt, Michael J. Freedman, Michael Kaminsky, and David G. Andersen. 2011. ”Don’t Settle for Eventual: Scalable Causal Consistency for Wide-Area Storage with COPS. ” In *Proceedings of the 23rd ACM Symposium on Operating Systems Principles*, 401-416. https://doi.org/10.1145/2043556.2043593

    work page doi:10.1145/2043556.2043593 2011

  22. [23]

    Liu, Tianyang, Canwen Xu, and Julian McAuley. 2024. ”RepoBench: Benchmarking Repository-Level Code Auto-Completion Sys- tems. ” In *Proceedings of the 12th International Conference on Learning Representations (ICLR 2024)*. https://doi.org/10 .48550/arXiv.2306.03091

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  23. [24]

    Ding, Yangruibo, Zijian Wang, Wasi Uddin Ahmad, Hantian Ding, Ming Tan, Nihal Jain, Murali Krishna Ramanathan, Ramesh Nallapati, Parminder Bhatia, Dan Roth, and Bing Xiang. 2023. ”CrossCodeEval: A Diverse and Multilingual Benchmark for Cross- File Code Completion. ” In *Advances in Neural Information Processing Systems 36*. arXiv:2310.11248. https://doi.o...

    work page arXiv 2023

  24. [25]

    Li, Wei, Xin Zhang, Zhongxin Guo, Shaoguang Mao, Wen Luo, Guangyue Peng, Yangyu Huang, Houfeng Wang, and Scarlett Li

  25. [26]

    ” In *Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics*, 17160–17176

    ”FEA-Bench: A Benchmark for Evaluating Repository-Level Code Generation for Feature Implementation. ” In *Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics*, 17160–17176. https://doi.org/10.48550/a rXiv.2503.06680

    work page doi:10.48550/a

  26. [27]

    Zan, Daoguang, Ailun Yu, Wei Liu, Dong Chen, Bo Shen, Wei Li, Yafen Yao, Yongshun Gong, Xiaolin Chen, Bei Guan, Zhiguang 39 Yang, Yongji Wang, Qianxiang Wang, and Lizhen Cui. 2024. ”CodeS: Natural Language to Code Repository via Multi-Layer Sketch. ” arXiv:2403.16443 [cs.LG]. https://doi.org/10.48550/arXiv.2403.16443

    work page doi:10.48550/arxiv.2403.16443 2024

  27. [28]

    Ding, Jingzhe, Shengda Long, Changxin Pu, Huan Zhou, Hongwan Gao, Xiang Gao, Chao He, Yue Hou, Fei Hu, Zhaojian Li, Weiran Shi, Zaiyuan Wang, Daoguang Zan, Chenchen Zhang, Xiaoxu Zhang, Qizhi Chen, Xianfu Cheng, Bo Deng, Qingshui Gu, Kai Hua, Juntao Lin, Pai Liu, Mingchen Li, Xuanguang Pan, Zifan Peng, Yujia Qin, Yong Shan, Zhewen Tan, Weihao Xie, Zihan W...

    work page arXiv 2025

  28. [29]

    Sun, Chengzheng, Xiaohua Jia, Yanchun Zhang, Yun Yang, and David Chen. 1998. ”Achieving Convergence, Causality Preservation, and Intention Preservation in Real-Time Cooperative Editing Systems. ” *ACM Transactions on Computer-Human Interaction* 5 (1): 63-108. https://doi.org/10.1145/274444.274447

    work page doi:10.1145/274444.274447 1998

  29. [30]

    Chacon, Scott, and Ben Straub. 2014. *Pro Git*, 2nd ed. Apress / Open Source. https://git-scm.com/book

    2014

  30. [31]

    Bernstein, Philip A., Vassos Hadzilacos, and Nathan Goodman. 1987. *Concurrency Control and Recovery in Database Systems*. Reading, MA: Addison-Wesley. https://www.microsoft.com/en-us/research/people/philbe/book/

    1987

  31. [32]

    Hou, Xinyi, Yanjie Zhao, Yue Liu, Zhou Yang, Kailong Wang, Li Li, Xiapu Luo, David Lo, John Grundy, and Haoyu Wang. 2024. ”Large Language Models for Software Engineering: A Systematic Literature Review. ” *ACM Transactions on Software Engineering and Methodology* 33 (8): 1-79. https://doi.org/10.1145/3695988

    work page doi:10.1145/3695988 2024

  32. [33]

    Chiu, Claire Cardie, Matthias Gallé, and Alexander M

    Zhao, Wenting, Nan Jiang, Celine Lee, Justin T. Chiu, Claire Cardie, Matthias Gallé, and Alexander M. Rush. 2025. ”Commit0: Library Generation from Scratch. ” In *Proceedings of the 13th International Conference on Learning Representations (ICLR)*. arXiv:2412.01769 [cs.SE]. https://doi.org/10.48550/arXiv.2412.01769

    work page doi:10.48550/arxiv.2412.01769 2025

  33. [34]

    Zhou, Qixing, Jiacheng Zhang, Haiyang Wang, Rui Hao, Jiahe Wang, Minghao Han, Yuxue Yang, Shuzhe Wu, Feiyang Pan, Lue Fan, Dandan Tu, and Zhaoxiang Zhang. 2026. ”FeatureBench: Benchmarking Agentic Coding for Complex Feature Development. ” arXiv:2602.10975 [cs.SE]. https://doi.org/10.48550/arXiv.2602.10975

    work page doi:10.48550/arxiv.2602.10975 2026

  34. [35]

    Ni, Ziyi, Huacan Wang, Shuo Zhang, Shuo Lu, Ziyang He, Wang You, Zhenheng Tang, Yuntao Du, Bill Sun, Hongzhang Liu, Sen Hu, Ronghao Chen, Bo Li, Xin Li, Chen Hu, Binxing Jiao, Daxin Jiang, and Pin Lyu. 2025. ”GitTaskBench: A Benchmark for Code Agents Solving Real-World Tasks Through Code Repository Leveraging. ” arXiv:2508.18993 [cs.SE]. https://doi.org/1...

    work page arXiv 2025

  35. [36]

    SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering

    Yang, John, Carlos E. Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik Narasimhan, and Ofir Press. 2024. ”SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering. ” In *Advances in Neural Information Processing Systems 37*. arXiv:2405.15793. https://doi.org/10.48550/arXiv.2405.15793

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2405.15793 2024

  36. [37]

    AgentSpec: Customizable Runtime Enforcement for Safe and Reliable LLM Agents

    Wang, Haoyu, Christopher M. Poskitt, and Jun Sun. 2025. ”AgentSpec: Customizable Runtime Enforcement for Safe and Reliable LLM Agents. ” arXiv:2503.18666 [cs.AI].https://doi.org/10.48550/arXiv.2503.18666

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2503.18666 2025

  37. [38]

    Zhao, Wei, Zhe Li, Peixin Zhang, and Jun Sun. 2026. ”ClawGuard: A Runtime Security Framework for Tool-Augmented LLM Agents Against Indirect Prompt Injection. ” arXiv:2604.11790 [cs.CR]. https://doi.org/10.48550/arXiv.2604.11790

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.11790 2026

  38. [39]

    Winston, Cailin, Claris Winston, and René Just. 2026. ”Solver-Aided Verification of Policy Compliance in Tool-Augmented LLM Agents. ” arXiv:2603.20449 [cs.SE].https://doi.org/10.48550/arXiv.2603.20449

    work page doi:10.48550/arxiv.2603.20449 2026

  39. [40]

    Sousa, Marcelo, Isil Dillig, and Shuvendu K. Lahiri. 2018. ”Verifying Semantic Conflict-Freedom in Three-Way Program Merges. ” arXiv:1802.06551 [cs.PL]. https://doi.org/10.48550/arXiv.1802.06551

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1802.06551 2018

  40. [41]

    Cavalcanti, Guilherme, Paulo Borba, Leonardo dos Anjos, and Jonatas Clementino. 2024. ”Semistructured Merge with Language- Specific Syntactic Separators. ” arXiv:2407.18888 [cs.SE]. https://doi.org/10.48550/arXiv.2407.18888

    work page doi:10.48550/arxiv.2407.18888 2024

  41. [42]

    Mohammadi, Bardia, Nearchos Potamitis, Lars Klein, Akhil Arora, and Laurent Bindschaedler. 2026. ”Atomix: Timely, Transactional Tool Use for Reliable Agentic Workflows. ” arXiv:2602.14849 [cs.LG].https://doi.org/10.48550/arXiv.2602.14849

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2602.14849 2026

  42. [43]

    Chen, Zheng, Hanqing Liu, Duling Xu, Dong Dong, Jialin Li, Bangzheng Pu, and Jidong Zhai. 2026. ”Cordon: Semantic Transactions for Tool-Using LLM Agents. ” arXiv:2606.17573 [cs.OS].https://doi.org/10.48550/arXiv.2606.17573

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2606.17573 2026

  43. [45]

    Mao, Zhenyu, Jacky Keung, Fengji Zhang, Shuo Liu, Yifei Wang, and Jialong Li. 2025. ”Towards Engineering Multi-Agent LLMs: A Protocol-Driven Approach. ” arXiv:2510.12120 [cs.SE].https://doi.org/10.48550/arXiv.2510.12120

    work page doi:10.48550/arxiv.2510.12120 2025

  44. [46]

    Hou, Bo, Xin Tan, Kai Zheng, Fang Liu, Yinghao Zhu, and Li Zhang. 2025. ”LLM-Driven Collaborative Model for Untangling Commits via Explicit and Implicit Dependency Reasoning. ” arXiv:2507.16395 [cs.AI]. https://doi.org/10.48550/arXiv.2 507.16395. 40

    work page doi:10.48550/arxiv.2 2025