Topology-Consistent Task Planning over Cellular Workflow Complexes for LLM-based Agents

2026-10-04

AI Summary

This study addresses the limitation of existing LLM agents in handling complex workflows involving verification-correction loops, branching-merging, and state reuse. To this end, we propose TopoPlanner, a framework that elevates tool dependency graphs to cell complexes to enhance nonlinear modeling capacity. TopoPlanner introduces a novel topology-aware retrieval mechanism based on cellular coherence, combined with multidimensional structural reasoning to assist LLMs in generating tool sequences. Evaluated across four benchmarks, TopoPlanner significantly outperforms prompt engineering and graph-augmented baselines on workflows containing loops and merges, effectively overcoming planning bottlenecks in complex workflow scenarios.

Abstract

Task planning for LLM agents requires workflows that satisfy both user intent and complex sub-task dependencies. While existing planners work well for sequential or directed acyclic graph (DAG)-like structures, they struggle with workflow patterns such as verification-correction loops, convergent branch merging, and reusable intermediate states that arise naturally in real-world tool orchestration. We present TopoPlanner, a topology-consistent planning framework that lifts tool dependency graphs into cellular workflow complexes and uses them as topologyaware context for LLM tool planning. TopoPlanner retrieves a request-relevant closed subcomplex through cosheaf-consistent cellular retrieval, performs multidimensional structural reasoning over the retrieved topology, and interfaces the resulting cellular representation with the planner LLM for tool-sequence generation. Experiments on four tool-planning benchmarks with topology-guided loop, merge, and loop-merge workflows show consistent improvements over prompt-based and graph-enhanced baselines across different local LLM backbones.

Problem

• Task Planning
• LLM-based Agents
• Workflow Complexes
• Tool Orchestration
• Sub-task Dependencies

Innovation

• Cellular Workflow Complexes
• Topology-Consistent Planning
• Cosheaf-Consistent Retrieval
• Task Planning
• LLM Agents