TAPE: Tool-Guided Adaptive Planning and Constrained Execution in Language Model Agents

2026-02-23

AI Summary

This work addresses the challenge of irreversible failures in language model agents operating under strong constraints, which often stem from flawed planning and stochastic execution. To mitigate this, the paper proposes a novel framework integrating tool-guided adaptive planning with constrained execution, uniquely combining external solvers and dynamic replanning mechanisms. The approach constructs multi-plan graphs, leverages external solvers to generate feasible action sequences, enforces constraint-aware decoding, and triggers adaptive replanning based on environmental feedback. This methodology substantially enhances agent robustness and task success rates, achieving an average improvement of 21.0 percentage points across challenging benchmarks—including Sokoban, ALFWorld, MuSiQue, and GSM8K-Hard—and yields gains of approximately 20.0 percentage points even when applied to weaker base models.

Technology Category

• Multiagent Systems: Multiagent Planning
• Planning, Routing, and Scheduling: Planning with Language Models
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Language Model (LM) agents have demonstrated remarkable capabilities in solving tasks that require multiple interactions with the environment. However, they remain vulnerable in environments where a single error often leads to irrecoverable failure, particularly under strict feasibility constraints. We systematically analyze existing agent frameworks, identifying imperfect planning and stochastic execution as the primary causes. To address these challenges, we propose Tool-guided Adaptive Planning with constrained Execution (TAPE). TAPE enhances planning capability by aggregating multiple plans into a graph and employing an external solver to identify a feasible path. During execution, TAPE employs constrained decoding to reduce sampling noise, while adaptively re-planning whenever environmental feedback deviates from the intended state. Experiments across Sokoban, ALFWorld, MuSiQue, and GSM8K-Hard demonstrate that TAPE consistently outperforms existing frameworks, with particularly large gains on hard settings, improving success rates by 21.0 percentage points on hard settings on average, and by 20.0 percentage points for weaker base models on average. Code and data available at here.

Problem

• Language Model Agents
• Feasibility Constraints
• Planning
• Execution Robustness
• Irrecoverable Failure

Innovation

• adaptive planning
• constrained execution
• tool-guided reasoning
• language model agents
• feasibility-aware planning