Exploration Structure in LLM Agents for Multi-File Change Localization

2026-06-10

AI Summary

This work addresses the inefficiency of existing large language model (LLM) agents in locating multi-file code changes, which stems from their reliance on linear exploration strategies that struggle with complex, cross-subsystem modifications. To overcome this limitation, the authors propose a domain-partitioned parallel agent architecture that enables nonlinear, concurrent exploration through the introduction of domain scoping. They develop a GitHub issue evaluation framework based on persistent sessions and employ lightweight Haiku-like models to facilitate collaborative exploration among domain-specific agents. Experimental results demonstrate that the proposed approach significantly outperforms comparable small models on an expanded benchmark, achieving the highest micro F1 score and overall performance second only to the substantially larger Codex 5.5 High model.

Abstract

Software engineering tools increasingly rely on LLM based agents to localize files to change to resolve a software issue. Most AI agents explore repositories linearly, that is, visiting one directory or file per step. We postulate that this is a structural mismatch for changes that span several subsystems. We compare linear sequential exploration against non-linear, domain-scoped parallel agentic exploration. Using SWE Bench Pro as initial benchmark, we focus on ansible as an exemplar. We construct an approach for persistent-session evaluation of GitHub issues anchored at a single base commit. We compare our non-linear domain-agent file traversal system against a base LLM without direct repository access, a single agent Recursive Language Model (RLM) baseline with a persistent Python REPL and an external CLI baseline using Codex 5.5 High. Domain scoped parallel agent spawning with a small Haiku-class model achieves the highest micro F1 among Haiku class models by a large margin. Domain-agents is the second highest behind only the much larger Codex 5.5 High on our own expanded benchmark including over more recent PRs from 2025 and 2026. On the original, curated, 2020 SWE-bench Pro benchmark, a larger Sonnet plain LLM baseline attains higher micro F1 by predicting few files, leading to higher precision, but at significantly lower all gold recall. We also present three additional findings. First, documentation evolution is a latent dependency unresolved by any approach. Second, naive file system access can degrade localization driven by test-file over prediction. Lastly, forced multi-agent consultation does not measurably help and raises token cost substantially.

Problem

• multi-file change localization
• LLM agents
• repository exploration
• software engineering
• structural mismatch

Innovation

• non-linear exploration
• domain-scoped agents
• multi-file change localization
• persistent-session evaluation
• parallel agentic traversal