coding-guidelines

To stop LLM coding mistakes, enforce explicit assumptions, simplicity, and surgical edits that only touch necessary lines. This approach biases toward caution over speed and structures implementation tasks around verifiable success criteria.

The best way to perform surgical code edits is to limit changes to exactly what is requested, preserve existing code style, and clean up only artifacts introduced by the current modification. This prevents overengineering during refactoring and bug fixes.

To force an LLM to think before coding, apply guidelines that require the model to explicitly state assumptions, uncertainties, and tradeoffs. This prevents silent guessing and ensures disciplined, goal-driven execution for feature development.

No, these LLM coding guidelines are not designed for architecture design or documentation. They specifically target implementation tasks such as feature development, refactoring, bug fixing, and test-oriented iteration to ensure safer code execution.

Unnecessary changes during bug fixes occur when LLMs lack constraints for surgical edits. By enforcing simplicity first and limiting output to what is actually requested, you can prevent overengineering and ensure edits touch only what is necessary.

Yes, test-and-verify loops are essential for goal-driven code execution. They reframe implementation tasks into verifiable success criteria, allowing you to validate changed lines and ensure the LLM's modifications map directly to your request.