AI Agent Recovers Missing Links Between Software Issues and Code Commits

Connecting software issues to the commits that fix them is vital for maintaining project records and understanding code evolution, yet it remains poorly solved. Evidence from GitHub repositories suggests that fewer than half of all issues receive proper commit linkage, creating significant gaps in traceability.

The Challenge of Current Approaches#

Traditional machine learning and deep learning models have tackled this problem with limited success. More recent efforts have turned to large language models, but they hit a wall: LLMs operate within fixed token budgets and struggle to process the sprawl of commit histories, issue discussions, and code files simultaneously.

Beyond these technical constraints lies a more fundamental issue. Most existing systems evaluate individual issue-commit pairs in isolation, treating the problem as a simple one-to-one matching exercise. This fails to capture how real software fixes work—where a single issue is typically resolved through a chain of related commits, not a single change. By missing these dependencies and temporal sequences, systems often misidentify intermediate changes as the actual fix and waste computational resources scoring countless irrelevant pairs.

LinkAnchor: A New Architecture#

Researchers have developed LinkAnchor, an autonomous LLM-based agent built specifically for this task. Its key innovation is a lazy-access architecture that enables the underlying model to fetch only the most pertinent information on demand—specific commits, comments on issues, and relevant code snippets—without breaching token limits.

This design sidesteps the dual constraints that have hampered prior work: it avoids overwhelming the LLM with irrelevant context, and it enables the agent to reason across chains of commits rather than evaluating them in isolation. The result is a system capable of understanding the full resolution logic embedded in how developers actually fix problems.

Implications for Software Development#

The recovery of issue-to-commit links matters far beyond academic interest. Accurate traceability underpins project management, regulatory compliance, code review workflows, and the ability to understand why specific changes were made. Improving this foundational capability could strengthen how teams track and communicate about their work.