AgenticSZZ: Temporal Knowledge Graph-Guided Agentic Bug-Inducing Commit Identification

Bug-Inducing Commit Identification: Identifying bug-inducing commits is important for understanding software defects and downstream software engineering tasks.

SZZ Blame Search Limitation: Existing SZZ-based approaches are limited because git blame restricts the search space to commits that directly modified fixed lines.

Beyond-Blame BIC Cases: A substantial portion of bug-inducing commit cases cannot be solved by blame alone because they require deeper history traversal or are blameless.

Bug-Inducing Commit Identification: Identifying bug-inducing commits is important for understanding software defects and downstream software engineering tasks.

SZZ Blame Search Limitation: Existing SZZ-based approaches are limited because git blame restricts the search space to commits that directly modified fixed lines.

AgenticSZZ: AgenticSZZ identifies bug-inducing commits by reframing the task from ranking blame commits into a temporal knowledge graph search problem.

Temporal Knowledge Graph for Software Evolution: The approach constructs a temporal knowledge graph encoding commits with temporal and structural relationships and expands the search space through backward file-history traversal from blame commits and the bug-fixing commit.

LLM Agent Graph Navigation: The approach uses an LLM agent with specialized tools to navigate the graph for candidate exploration and causal analysis.

AgenticSZZ: AgenticSZZ identifies bug-inducing commits by reframing the task from ranking blame commits into a temporal knowledge graph search problem.

Temporal Knowledge Graph for Software Evolution: The approach constructs a temporal knowledge graph encoding commits with temporal and structural relationships and expands the search space through backward file-history traversal from blame commits and the bug-fixing commit.

F1 Improvement over State of the Art:

Component Necessity Confirmed:

New Research Direction: