Поиск антипаттернов производительности в Java

engineering

performance-smell-detection is a Claude Code skill that identifies potential code-level performance smells in Java: string concatenation in loops, Stream creation inside tight loops, boxing/unboxing in hot paths, per-iteration Regex compilation, and unbounded collection growth. It accounts for modern JVM behavior — Java 9+ optimizes simple `+` concatenation via `invokedynamic`, so the skill focuses only on patterns that remain genuinely problematic in Java 21/25. Each smell is rated by severity (High, Medium, or Low) with context explaining when a fix is warranted and when readability should take priority. The guiding principle is measure first — with JMH, a profiler, or production metrics — before any optimization. For JPA and database performance issues, the `jpa-patterns` skill is recommended instead.