Engineering a Governance-Aware AI Sandbox: Design, Implementation, and Lessons Learned

Collection of recent software-engineering and AI-for-code papers. Key contributions: a governance-aware, multi-tenant AI sandbox architecture for traceable, auditable experiments; a dual-model interaction pattern (coder + reviewer) that significantly improves code synthesis; a study linking semantic neighborhood density to programmer eye gaze; CONCUR, a benchmark for LLM-generated concurrent code (deadlocks/races); SWE-CI, a repository-level CI benchmark to evaluate long-term maintainability of LLM agents; methodology for running adversarial reverse-engineering studies with students; LoRA-MME,