Claude Code Claude Opus 5.5 uses ten-pass review
The author explains a structured method for reviewing AI-generated code by running ten focused prompt passes sequentially in an agent conversation. Each narrow pass checks a single concern, such as architecture, security, or performance, until no confirmed issues remain. While this technique significantly improves code quality, the author notes that it still misses library-level flaws and incorrect requirements that must be caught by tests and outside reviews. Open article →