LLM Code Smells: A Taxonomy and Detection Approach

This feed aggregates multiple 2026 CS papers with strong relevance to software security and reliability. Key items: a taxonomy of nine "LLM code smells" plus SpecDetect4LLM (static detector) reporting 91.3% precision, 71.8% recall and LLM-smell prevalence in 73.5% of 692 OSS systems; an empirical study showing seven popular LLMs produce vulnerable code with the majority of vulnerabilities rated high or critical; QueryZen/Branch Flip Analysis which uncovered 21 previously unknown performance issues in major DBMSs by systematically flipping optimization branches; agent-breakage, an open-source,闭