Linter-Guided Repair of Quantum Code Smells

Quantum programs can suffer from quantum-specific code smells that cannot be captured by classical unit tests. We propose a framework that analyzes the existing programs released in two popular benchmarks using LintQ, a linter for quantum code smells, and attempts to repair them using LLMs while leaving functional correctness untouched. We test the proposed pipeline on 14 different LLMs and 2 popular Qiskit benchmarks and the results show that existing benchmarks are not free from code smells but repairing them using LLMs is feasible without breaking their functional correctness.

To appear in SaTQuML: Secure and Trustworthy Quantum Machine Learning Workshop, NeurIPS 2026