A DevOps team has wired Claude Code into their CI/CD pipeline. It reviews PRs, generates tests, runs security scans, and checks deployments across a polyglot codebase with Terraform underneath.
A CI/CD system prompt defines two review categories with the instructions 'Check for security vulnerabilities in each function' and 'Check for performance issues in each loop'. The model frequently calls `performance_check` for security issues found inside loops, and `security_check` for performance issues in security-sensitive functions. What is the root cause and best fix?
live mock