mlsanityData to 7 Oct 2026

Control-Data Flow Separation: Stable Prompt Optimization in Multi-Agent LLMs

arXiv 2609.00621 ↗First posted 1 Sep 2026cs.AI cs.CL cs.MAcited by 0

On the map: LLM agents and tool use → Software engineering coding agents

Abstract (arXiv)

Prompt optimization can improve multi-agent LLM systems, but the prompts being optimized often serve two entangled roles: generating task-relevant content and specifying execution-critical protocols, such as message routing, output formatting, and termination signals, on which the underlying code relies. As a result, a prompt edit intended to improve content generation can inadvertently corrupt the protocol and cause the entire agent pipeline to fail. Our key observation is that these two roles have different representations: execution protocols are typically structured, while task-relevant content is usually expressed in unstructured language. Based on this, we propose control-data flow separation, where execution-critical control is represented as typed, validated program objects, while task-relevant language remains the optimizable data flow for agent communication. This design allows optimizers to improve multi-agent behavior without exposing the routing or formatting interface to prompt drift. Across synthetic reasoning, collaborative review generation, and insurance rating workflows, our framework empirically achieves 100% eventual protocol validity while consistently improving task performance.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use. Builds on 33 of its 52 references that are in the map. First posted on arXiv 1 Sep 2026.

Builds on

33 of 52 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024SWE-Search: Enhancing Software Agents with Monte Carlo Tree Search and Iterative Refinement26 Oct
  2. 2024Agentless: Demystifying LLM-based Software Engineering Agents1 Jul
  3. 2023AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation16 Aug
  4. 2023ChatDev: Communicative Agents for Software Development16 Jul

Other references · 29

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  2. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,317
  3. 2023G-Eval: NLG Evaluation using GPT-4 with Better Human Alignmentcited by 2,205
  4. 2022Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Themcited by 1,767
  5. 2023Improving Factuality and Reasoning in Language Models through Multiagent Debatecited by 1,739
  6. 2024A Survey on LLM-as-a-Judgecited by 1,384
  7. 2023CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Societycited by 1,348
  8. 2023AgentBench: Evaluating LLMs as Agentscited by 1,211
  9. 2024Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preferencecited by 1,151
  10. 2022Large Language Models Are Human-Level Prompt Engineerscited by 1,020
  11. 2024The AI Scientist: Towards Fully Automated Open-Ended Scientific Discoverycited by 1,002
  12. 2024Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluatorscited by 865
17 more
  1. 2023Automatic Prompt Optimization with "Gradient Descent" and Beam Searchcited by 628
  2. 2023Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolutioncited by 521
  3. 2021PICARD: Parsing Incrementally for Constrained Auto-Regressive Decoding from Language Modelscited by 402
  4. 2024Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programscited by 270
  5. 2023Unleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaborationcited by 246
  6. 2023PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimizationcited by 226
  7. 2023Can large language models provide useful feedback on research papers? A large-scale empirical analysiscited by 195
  8. 2023Grammar-Constrained Decoding for Structured NLP Tasks without Finetuningcited by 186
  9. 2023ReviewerGPT? An Exploratory Study on Using Large Language Models for Paper Reviewingcited by 118
  10. 2023GoLLIE: Annotation Guidelines improve Zero-Shot Information-Extractioncited by 116
  11. 2024MARG: Multi-Agent Review Generation for Scientific Paperscited by 98
  12. 2024CodeR: Issue Resolving with Multi-Agent and Task Graphscited by 89
  13. 2024SynCode: LLM Generation with Grammar Augmentationcited by 66
  14. 2023ARIES: A Corpus of Scientific Paper Edits Made in Response to Peer Reviewscited by 34
  15. 2024Symbolic Prompt Program Search: A Structure-Aware Approach to Efficient Compile-Time Prompt Optimizationcited by 29
  16. 2025Thought Communication in Multiagent Collaborationcited by 28
  17. 2025MASTER: A Multi-Agent System with LLM Specialized MCTScited by 15

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026LLM-as-Code: Agentic Programming for Agent Harness
  2. 2026Utility-Guided Agent Orchestration for Efficient LLM Tool Use
  3. 2026MAS-PromptBench: When Does Prompt Optimization Improve Multi-Agent LLM Systems?
  4. 2025A Comprehensive Survey on Benchmarks and Solutions in Software Engineering of LLM-Empowered Agentic System
  5. 2026Enhancing LLM Instruction Following: An Evaluation-Driven Multi-Agentic Workflow for Prompt Instructions Optimization
  6. 2025CodeAgents: A Token-Efficient Framework for Codified Multi-Agent Reasoning in LLMs
  7. 2026PerspectiveGap: A Benchmark for Multi-Agent Orchestration Prompting
  8. 2026Learning What to Skip: Counterfactual Credit Assignment for Efficient Multi-Agent LLM Workflows