A Local Human‑in‑the‑Loop Assistant for AI‑Assisted SVN Commit‑Message Generation

This technical report presents svn_ai_msg, an open-source command-line assistant for generating commit-message candidates from Subversion working copies. The tool combines deterministic repository inspection, lightweight rule-based processing, and local Large Language Model inference through Ollama. The proposed workflow extracts information from \texttt{svn status} and selected diff content, generates multiple commit-message candidates, and preserves user control over the final commit operation. The implementation is designed for local execution in order to support source-code confidentiality, reproducibility, and integration with lightweight command-line environments. A preliminary functional evaluation on representative SVN changes indicates that the proposed workflow can generate useful commit-message candidates while maintaining a human-in-the-loop interaction model. The source code is publicly available and archived with a persistent DOI.