StructAgent: Harness Long-horizon Digital Agents with Unified Causal Structure

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Wenyi Wu, Sibo Zhu, Kun Zhou, Aayush Salvi, Zixuan Song, Biwei Huang
• Venue: arXiv: cs.AI
• Published: 2026-07-13
• Collected: 2026-07-14
• URL: http://arxiv.org/abs/2607.11388v1
• PDF: https://arxiv.org/pdf/2607.11388v1
• DOI: UNKNOWN

Plain-Language Brief

In this paper, researchers introduce StructAgent, a framework designed to improve long-horizon digital agents' performance in tasks with evolving contexts and complex workflows. The framework introduces a unified state for maintaining task progress and a structured workflow for regulating progress through verifier-backed state transitions. It also allows for explicit progress checkpointing, evidence-driven task completion, targeted failure recovery, and tool-supported execution, ensuring all progress updates are grounded in verification.

What the Document Says

• Recent advances have led to increasingly capable digital agents for computer use.
• Long-horizon tasks involve evolving contexts with accumulated observations, intermediate edits, failed attempts, and partially completed executions.
• Existing agents operate over raw interaction history, making task progress difficult to interpret, verify, and recover.
• The paper argues that a solution requires structuring the agent's state and workflow around a unified causal representation of task progress.
• StructAgent is introduced, a framework with a state-centered design.
• It includes a unified state for maintaining compact, verifiable task progress and a structured workflow that regulates progress through verifier-backed state transitions.
• The framework allows for explicit progress checkpointing, evidence-driven task completion, targeted failure recovery, and tool-supported execution.
• All progress updates are grounded in verification within StructAgent.
• Extensive experiments show that StructAgent improves various LLM and VLM backbones on long-horizon computer-use tasks.
• On OSWorld-Verified, it significantly increases the success rate for Qwen3.5-9B and Qwen3.5-27B, achieving a new open-source state of the art of 78.9% with MiniMax-M3.

Key Concepts Mentioned

• digital agents
• long-horizon tasks
• evolving contexts
• unified causal representation
• StructAgent
• state-centered framework
• verifier-backed state transitions
• explicit progress checkpointing
• evidence-driven task completion
• targeted failure recovery
• tool-supported execution

Claims Made by the Document

• Existing agents struggle with long-horizon tasks due to raw interaction history.
• StructAgent improves various LLM and VLM backbones on long-horizon computer-use tasks.
• StructAgent significantly increases the success rate for Qwen3.5-9B and Qwen3.5-27B on OSWorld-Verified.

Evidence Used

• Improvements demonstrated through extensive experiments

Methods / Approach

• Introduction of a unified state and structured workflow in StructAgent

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

The text does not explicitly mention any limitations or potential issues with the proposed approach.

Open Questions

• What specific techniques are used within StructAgent for progress checkpointing, evidence-driven task completion, targeted failure recovery, and tool-supported execution?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The text does not provide detailed information on how these aspects are achieved within the framework.

Raw Abstract / Excerpt

Recent advances in large language models (LLMs) and vision-language models (VLMs) have enabled increasingly capable digital agents for computer use. However, real-world tasks are often long-horizon and involve evolving contexts containing accumulated observations, intermediate edits, failed attempts, and partially completed executions. Existing agents typically operate over raw interaction history, making task progress difficult to interpret, verify, and recover, which ultimately limits reliable long-horizon execution. In this paper, we argue that addressing this challenge requires explicitly structuring both the agent's state and workflow around a unified causal representation of task progress. We present \textbf{StructAgent}, a state-centered framework that introduces a unified state for maintaining compact, verifiable task progress and a structured workflow that regulates progress through verifier-backed state transitions. Building on this design, StructAgent further enables explicit progress checkpointing, evidence-driven task completion, targeted failure recovery, and tool-supported execution, while ensuring that all progress updates remain grounded in verification. Extensive experiments demonstrate that StructAgent consistently improves a wide range of LLM and VLM backbones on long-horizon computer-use tasks. On OSWorld-Verified, it improves Qwen3.5-9B from 27.0% to 46.9% success rate and Qwen3.5-27B from 31.6% to 62.2%, while achieving a new open-source state of the art of 78.9% with MiniMax-M3. Moreover, the same framework generalizes beyond desktop environments to Minecraft, demonstrating the generality of our design.