Holmes: Multimodal Agentic Diagnosis for Mixed-Language Mobile Crashes at Industrial Scale

2026-06-20

AI Summary

This work addresses the significant challenges of crash diagnosis in ultra-large-scale industrial mobile applications operating in mixed-language environments, where code complexity and non-reproducible failures hinder effective debugging. The authors propose Holmes, a novel multi-agent system that automatically reconstructs failure contexts and identifies root causes without requiring environment reproduction by innovatively integrating multimodal runtime signals—such as stack traces, logs, and thread states. Holmes employs a hierarchical retrieve-explore-reason architecture, leveraging low-level information like registers and assembly instructions to bridge the semantic gap between open-source application logic and closed-source system frameworks. Guided by runtime clues, it dynamically narrows the search space through intelligent code navigation. Evaluated on real-world crash data from WeChat, Holmes achieves an 87.6% function-level localization accuracy and reduces average diagnosis time by 98%, down to approximately 77 seconds.

Abstract

Diagnosing mobile crashes in ultra-large-scale industrial applications is a formidable challenge due to the sheer volume of code, the complexity of mixed-language environments, and the inability to reproduce failures locally. Traditional static analysis struggles with scalability, while existing LLM-based agents often rely on reproducible environments unavailable in post-mortem scenarios. We present Holmes, a multi-agent system that automates root cause analysis by synthesizing multimodal runtime signals--stack traces, logs, and thread states--to reconstruct failure contexts without reproduction. Holmes introduces a hierarchical Retrieve-Explore-Reason architecture that leverages low-level artifacts (e.g., registers, assembly) to bridge the semantic gap between open-source business logic and closed-source system frameworks. By dynamically compressing the search space using runtime clues, Holmes precisely navigates 70-million-line codebases to identify non-local defects. Evaluated on real-world crashes from WeChat, Holmes achieves 87.6% accuracy in function-level fault localization and reduces average investigation time by over 98% (to ~77 seconds), demonstrating its effectiveness in transforming labor-intensive debugging into an efficient verification workflow.

Problem

• mobile crashes
• mixed-language environments
• industrial-scale diagnosis
• non-reproducible failures
• root cause analysis

Innovation

• multimodal agentic diagnosis
• mixed-language crash analysis
• hierarchical Retrieve-Explore-Reason
• runtime signal synthesis
• non-local defect localization