Agolic: AI Planning System Enhances Symbolic Execution Coverage

A recent paper published on arXiv (2608.06397) presents Agolic, a planning system aimed at enhancing the effectiveness of symbolic execution. This technique analyzes programs by investigating possible execution paths, but practical applications often deplete resources before fully exploring the program's behavior. Agolic mitigates this issue by considering the tool's usage across several limited runs, while the symbolic execution tool manages the core state exploration. It leverages data from previous runs to optimize the configuration of subsequent bounded symbolic execution (BSE) runs. The system's planning capabilities, available information, and execution strategies can be tailored to fit specific symbolic execution tools and analysis goals. One adaptation for branch-coverage exploration uses an LLM-based agent to analyze source code, replay coverage, and prior targeting attempts. Evaluations on various C and C++ programs demonstrated enhanced coverage across all tested programs. This work holds significance for software engineering and artificial intelligence, especially in automated program analysis.