AI-powered security audit skill for code changes with false positive filtering. Based on Anthropic's claude-code-security-review.