Linter-Guided Repair of Quantum Code Smells

Quantum programs can suffer from quantum-specific code smells that cannot be captured by 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.