Diagnosing with Insights: Structured Analysis of Agent Failures via Behavioral Abstractions
Jiayi Bi (Tsinghua), Yanjie Gao and colleagues at Microsoft Research, with Tianyin Xu of UIUC, present AGENTSCOPE, a neuro-symbolic failure diagnosis system that abstracts long agent trajectories into structured behavioral representations and checks them against declarative 'neural invariants' to localize both the failing step and its failure type.
Ask this paper
Behavioral abstraction over raw traces: Rather than feeding a 100-step trajectory to a judge model, AGENTSCOPE compiles the trace into a structured representation, so the search for the failing step happens over program-like objects instead of prose.
Neural invariants as specifications: The paper borrows the invariant idea from program analysis and softens it: properties of agent behavior stated in natural language that an LLM checks against the abstraction. This is the bridge between symbolic rigor and the fuzziness of agent behavior.
AgentErrata, a new failure dataset: Beyond the existing Who&When benchmark, the authors release a broader agent-failure dataset and report state-of-the-art fault localization and attribution accuracy on both.
Why it matters: Every team running agents in production has the same problem: a run failed 80 steps ago and nobody knows where. Pure LLM-as-judge diagnosis is unreliable and pure static analysis does not apply. The structured-abstraction plus invariant-checking recipe is the most credible middle path published so far.
Abstract
With the proliferation of LLM agents, the ability to understand and diagnose failures in agents is essential to achieving superior effectiveness and trustworthiness. As agent failures often manifest via long and complex trajectories, manually finding the needles in the haystack is untenable. However, traditional diagnosis techniques for software bugs can hardly address LLM agent failures, while completely relying on LLMs as the judge yields unreliable diagnosis results. To overcome these challenges, this paper presents AGENTSCOPE, a new neuro-symbolic approach for agent failure mode diagnosis. The key principle of AGENTSCOPE is to abstract agent behavior, based on its trajectories, into structured representations. Furthermore, AGENTSCOPE introduces the concept of neural invariants to specify agent behavior properties. AGENTSCOPE leverages LLM-guided reasoning atop the structured representation against neural invariants to pinpoint both the failure step and its type in the trajectory. We show the effectiveness of AGENTSCOPE on publicly available agent failure datasets (Who&When) and a more comprehensive dataset created by us (AgentErrata), where AGENTSCOPE significantly outperforms the current state of the art in fault localization and attribution accuracy. Our work shows that integrating structured abstractions with LLM-guided reasoning enables effective, reliable, and interpretable diagnosis for agent failures.