Skip to content

How Do Agentic LLMs Decide to Call Tools? A Tool-Call Vector Shaped by Suppression

Oct 2026 · 0 citations · 60 references
Computer Science

TL;DR

This work proposes a method that converts complex agentic prompts into minimal contrastive pairs in which a single request verb determines the tool-call decision, and traces the decision to a vector, $\mu_\Delta$, that is both causally necessary and sufficient and generalizes beyond the discovery prompts to native multi-turn $\tau^2$-Bench trajectories and verb-free requests.

Abstract

Tool calling, invoking external tools on demand, is central to agentic LLMs, yet the mechanism that decides whether a model calls a tool or responds directly remains poorly understood. Agentic prompts are long and heavily scaffolded, combining role instructions, tool schemas, format templates, and the user's request across hundreds of tokens, creating a noisy, highly entangled context in which no single controllable variable for mechanistic analysis is obvious. To obtain such a variable, we propose a method that converts complex agentic prompts into minimal contrastive pairs in which a single request verb determines the tool-call decision: replacing an execution-verb (e.g., \textit{write}) with an analysis-verb (e.g., \textit{discuss}) reliably flips the decision, suggesting it is mediated by a compact internal state. We construct 500 such paired prompts across Python, Java, and C++ (300 for mechanistic analysis, 200 held out for evaluation). We trace the decision to a vector, $\mu_\Delta$, that is both causally necessary and sufficient and generalizes beyond the discovery prompts to native multi-turn $\tau^2$-Bench trajectories and verb-free requests. Behavioral ablations show that the scaffold establishes a tool-call prior; Transcoder decomposition then reveals that analysis verbs suppress this prior through features signaling that tool use is unnecessary, whereas execution verbs largely leave it intact. Downstream scaffold-reading attention heads and MLP features read out the resulting state, and the same mechanism recurs across seven models from the Qwen, Mistral, and Granite families. Our code is available at https://github.com/XijieGo/MI4ToolCalling.

View source

Similar papers

#artificial intelligence Preprint Oct 2026

Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files

Modern agent frameworks allow users to provide Markdown files containing system prompts, tool descriptions, and skill definitions (e.g., SKILL.md), which are incorporated into the agent's context during initialization. In this work, we investigate whether this configuration mechanism can also serve as an injection chan...

K. R. Manjusha, M. Emani · 0 citations
Preprint Oct 2026

Small agents should know what their tools do

Small open language models can call tools, but they do not reason about what a call does: they delete without asking, repeat writes, re-read state they already hold, and issue one call per turn where a short program would do. We introduce EffectWorld, an executable world of effect-typed tools in which every tool declar...

Alexander Kovalenko, Artem Redchych · 1 citation

Action-Space Shaping for LLM Agents: Measuring and Mitigating Tool-Schema Bias

Large Language Models (LLMs) have shown strong performance on tool-use agentic tasks when given a fixed tool schema. Yet a tool schema is not the action space of an agent; it is merely one interface representation of it. The same executable action can be exposed through many different, functionally equivalent tool defi...

Yinhong Liu, Zhi-Li Tan, Zi-Lin Wang et al. · 1 citation

AgentHop: A Diagnostic Benchmark for Agentic Multi-Hop Scientific Question Answering

AgentHop, a diagnostic benchmark of 1,011 multiple-choice questions paired with a controlled seven-tool sandbox under fixed token, turn, and tool-call constraints, and the full benchmark set and the harness are released to support diagnostic agent benchmarking.

Chanhee Park, Jeongho Yoon, Sung-Bin Han et al. · 0 citations
Sep 2026

Control-Token Injection Suppresses Chain-of-Thought and Defeats Reasoning-Based Oversight in Tool-Using Agents

The safety of a tool-using language model agent is usually treated as a property of the model alone. We give controlled, full-precision evidence that it is instead a joint property of the model and the software that renders its chat template and parses its tool calls, the decoding harness, and that both halves are atta...

Muhammad Usama, Khairatul Nisa, Summer Yeoreum Jung · 0 citations
#natural language process... Preprint Oct 2026

Before Retrieval Begins: Safety and Tool-Use Decisions in LLM Agents

Large language models increasingly act as agents that call external tools, yet most tool-safety research examines only what follows a tool call. The prior decision of whether to call the tool remains largely unexamined and can influence downstream safety. Across the seven open-weight models, the unweighted model-averag...

Aditya Nawal, Manit Baser, M. Gurusamy · 0 citations

Related blog posts

MIT News · Artificial Intelligence Sep 29, 2026

Who we become when we talk to machines

Professor Sherry Turkle’s new book, “Artificial Intimacy,” offers a withering critique of chatbots and the antisocial dynamics she believes they encourage.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.