How to Interpret Agent Behavior

TL;DR

ACT*ONOMY is a structured taxonomy and analysis tool for interpreting autonomous agent behavior from unstructured natural language trajectories, aiding diagnosis and oversight.

Contribution

It introduces a hierarchical taxonomy and an open repository for analyzing and comparing agent behaviors at runtime, facilitating better understanding and control.

Findings

ACT*ONOMY can compare behavioral profiles across different agents.

It characterizes an agent's behavior across diverse trajectories.

The taxonomy surfaces patterns indicative of failure modes.

Abstract

Autonomous agents such as Claude Code and Codex now operate for hours or even days. Understanding their runtime behavior has become critical for downstream tasks such as diagnosing inefficiencies, fixing bugs, and ensuring better oversight. A primary way to gain this understanding is analyzing the reasoning trajectories and execution traces these agents generate. Yet such data remains in unstructured natural-language form, making it difficult for humans to interpret at scale. We introduce ACT*ONOMY (a combination of Action and Taxonomy), a taxonomy for describing and analyzing agent behavior at runtime. ACT*ONOMY has two components: (1) the taxonomy itself, developed through Grounded Theory and structured as a three-level hierarchy of 10 actions, 46 subactions, and 120 leaf categories; and (2) an open repository that hosts the living taxonomy, provides an automated analysis pipeline that applies it to agent trajectories analysis, and defines an extension protocol for customization and growth. Our experiments show that ACTONOMY can compare behavioral profiles across agents and characterize a single agent's behavior across diverse trajectories, surfacing patterns indicative of failure modes. By providing a shared vocabulary, ACT*ONOMY helps researchers, agent designers, and end users interpret agent behavior more consistently, enabling better oversight and control.