ACCORD: Action-Conditioned Contextual Grounding for Language Agents

2026-06-15

AI Summary

This work addresses the challenge that large language model agents often misinterpret implicit context due to ambiguous instructions, leading to biased decision-making. To mitigate this issue, the authors propose ACCORD, a novel framework that, for the first time, integrates action-conditioned active environment probing with trajectory memory into language agents. ACCORD enables adaptive context grounding without requiring additional training or task-specific success signals. By dynamically querying the environment, retrieving relevant memories, and fusing contextual information, the framework substantially enhances the agent’s ability to perceive and leverage implicit cues. Experimental results demonstrate that ACCORD improves the task success rate of GPT-5-mini on AppWorld from 42.0% to 62.6% (+20.6 percentage points) and consistently yields performance gains across multiple benchmarks, including Claude-4.5-sonnet, Qwen3.5-27B-FP8, and AlfWorld.

Abstract

User instructions are often underspecified because humans rely on implicit assumptions about the surrounding environment. For large language model (LLM) agents operating in information-rich digital and physical environments, these assumptions cannot be inferred from the instruction alone; they must be recovered from the current state of tools, data, interfaces, and observations. Effective execution therefore requires agents to identify missing context, ground it in observed evidence, and carry it forward into subsequent actions. We show that current agents often fail to do so. They act from assumed rather than observed specifics, overlook information they could have gathered, and fail to incorporate evidence that has already been returned. Building on this insight, we propose ACCORD (Action-Conditioned Contextual Grounding), a simple and effective agent framework for adaptive grounding. Before each action, ACCORD actively probes the environment for missing information and integrates relevant context from the agent's trajectory that would otherwise be overlooked. Requiring no additional training or task-success signals, ACCORD improves task-goal completion on AppWorld by up to +20.6 points with GPT-5-mini, from 42.0% to 62.6%, compared to strong baselines. These gains persist with a substantially stronger base model (+10.8 with Claude-4.5-sonnet), an open-weight model (+10.1 with Qwen3.5-27B-FP8), and on the embodied AlfWorld benchmark (+7.4 success rate with GPT-5-mini).

Problem

• underspecified instructions
• contextual grounding
• language agents
• environmental context
• missing context

Innovation

• contextual grounding
• action-conditioned reasoning
• LLM agents
• environment probing
• trajectory-aware integration