XFlow: An Executable Protocol Programming System for Reliable Multi-Agent Workflows

2026-06-10

AI Summary

Large language model (LLM)-based multi-agent systems often suffer from insufficient reliability in coordinated planning, reasoning, and tool invocation due to ambiguous boundaries between prompting and execution. This work proposes XFlow, a novel system accompanied by the executable protocol language XPF, which introduces an executable protocol programming paradigm to establish a structured intermediate layer between pure prompt orchestration and token-based workflows. XPF explicitly encodes constraints, evidence handling, and procedural requirements through typed, verifiable lifecycle symbols, and integrates output mediation with protocol compilation to enforce clear commitment delineation and runtime guarantees. Experimental results demonstrate that this approach significantly enhances system reliability in settings involving constrained interactions, long-context reasoning, and agent-based software engineering tasks, effectively curbing error propagation and ensuring workflow consistency.

Abstract

LLM-based multi-agent systems increasingly coordinate planning, reasoning, tool use, and human interaction, yet their reliability remains limited. A central source of this limitation is the underspecified prompt--harness boundary. Current systems lack a principled way to decide which workflow commitments should remain in prompts and which should become harness structure. We present \textbf{XFlow}, an executable protocol programming system for reliable multi-agent workflows, and \textbf{XPF} (XFlow Protocol Format), its domain-specific protocol programming language. XFlow occupies a middle position between prompt-only orchestration and markup-like workflow descriptions. XPF remains readable as a literate protocol, but it is compiled and executed as a program. Its design keeps informal semantic work inside actors while moving selected commitments into harness structure that can be checked, preserved, and enforced. At runtime, XFlow stages uncertainty through lifecycle-governed symbols, which are typed state cells with validation and commit states. Actor outputs are mediated before they become shared state, instead of spreading through prompts, transcripts, or implicit memory. Our experiments cover Constrained Interaction, Long-Context Reasoning, and Agentic Software Engineering. They show that XFlow improves reliability by making constraints, evidence handling, and process requirements explicit and enforceable.

Problem

• multi-agent systems
• reliability
• prompt-harness boundary
• workflow commitments
• executable protocols

Innovation

• executable protocol
• multi-agent workflows
• protocol programming
• harness structure
• lifecycle-governed symbols