GIST-CMTF: Goal-State Inference for Causal Minimal Tool Filtering in LLM Agents

2026-06-15

AI Summary

This work addresses the challenge that user requests often contain multi-goal ambiguity, leading existing Causal Minimal Tool Filtering (CMTF) methods to execute incorrect goal paths. To resolve this, we propose the GIST-CMTF framework, which for the first time integrates goal-state reasoning into the CMTF pipeline. Specifically, it predicts candidate symbolic goals via a state-transition lexicon and employs an ambiguity-aware mechanism to dynamically decide whether to apply tool filtering or trigger a clarification action to recover missing information. While preserving the advantage of single-tool exposure, our approach substantially improves goal execution accuracy. Evaluated on 120 tasks, GIST-CMTF achieves a 97.0% success rate, reducing erroneous goal executions from 19.4% to 2.5%, significantly outperforming prior CMTF methods while saving substantial token usage.

Abstract

Tool-augmented LLM agents rely on runtime filtering to decide which tools should be visible at each step. Causal Minimal Tool Filtering (CMTF) reduces tool-choice confusion by exposing only the next causally necessary tool frontier, but it assumes that the user request has already been mapped to a symbolic goal state. In practice, requests such as "handle my appointment" or "take care of this email" may correspond to multiple possible goals. This creates wrong-goal execution, where an agent follows a valid causal tool path for an unintended objective. We introduce GIST-CMTF, a goal-state inference layer that predicts candidate symbolic goals over the same state-transition vocabulary used by CMTF, estimates ambiguity, and either applies CMTF or exposes clarification as a causal action that produces missing goal or state variables. We evaluate GIST-CMTF across seven model backends, six filtering methods, and 120 controlled tool-use tasks. GIST-CMTF achieves 97.0% task success, compared with 80.1% for top-goal CMTF and 82.9% for semantic-goal CMTF. It reduces wrong-goal execution from 19.4% under top-goal CMTF to 2.5%, while preserving the one-tool exposure of causal filtering and using substantially fewer tokens than all-tools exposure. These results suggest that reliable tool-augmented agents should validate goal state, not only tool relevance, before exposing external actions.

Problem

• goal-state inference
• tool-augmented LLM agents
• wrong-goal execution
• causal minimal tool filtering
• symbolic goal ambiguity

Innovation

• Goal-State Inference
• Causal Minimal Tool Filtering
• LLM Agents
• Tool Augmentation
• Ambiguity Resolution