To mitigate the inefficiencies of LLM-generated code—which often suffers from poor structure, "spaghetti" patterns, and repetitive styling errors—the author developed a specialized "agent.md" configuration file. By placing this file in a project's root directory, developers can inject persistent, granular instructions into an agentic IDE's prompt context, effectively automating the code review process. The implementation enforces strict technical standards: it mandates the use of enums over booleans, prohibits "magic numbers" by requiring descriptive constants, and enforces an architectural layered boundary hierarchy where components only communicate with immediate neighbors. To improve readability, the rules demand short function names (under 30 characters), the use of braces for all "if" statements, and the avoidance of the "arrow anti-pattern" through early returns. Additionally, the framework prescribes a specific seven-rule methodology for Git commit messages, emphasizing the imperative mood and a 50-character subject line limit. For bug fixing, the author mandates a test-driven approach: writing a failing test before implementing the fix. This method addresses "context dilution"—where models lose focus on middle-context instructions—by allowing the developer to instruct the agent to update the agent.md file itself, ensuring that style preferences remain at the boundaries of the prompt where attention is highest.