![[2602.19633] TAPE: Tool-Guided Adaptive Planning and Constrained Execution in Language Model Agents](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19633] TAPE: Tool-Guided Adaptive Planning and Constrained Execution in Language Model Agents
Summary
The paper presents TAPE, a novel framework for enhancing language model agents' planning and execution capabilities, addressing vulnerabilities in high-stakes environments.
Why It Matters
As language model agents become increasingly prevalent in complex tasks, improving their reliability under strict constraints is crucial. TAPE's approach to adaptive planning and execution could significantly enhance their performance, making them more viable for real-world applications.
Key Takeaways
- TAPE enhances planning by aggregating multiple plans into a graph.
- It uses an external solver to identify feasible execution paths.
- Constrained decoding minimizes sampling noise during execution.
- Adaptive re-planning occurs based on environmental feedback.
- TAPE shows significant performance improvements in challenging scenarios.
Computer Science > Artificial Intelligence arXiv:2602.19633 (cs) [Submitted on 23 Feb 2026] Title:TAPE: Tool-Guided Adaptive Planning and Constrained Execution in Language Model Agents Authors:Jongwon Jeong, Jungtaek Kim, Kangwook Lee View a PDF of the paper titled TAPE: Tool-Guided Adaptive Planning and Constrained Execution in Language Model Agents, by Jongwon Jeong and 2 other authors View PDF HTML (experimental) 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 percenta...