GPT Engineer CLI Cheatsheet - gpte Code Generation Full Reference

GPT Engineer sits in a different category from Copilot-style tools. It does not complete the function you are typing; it accepts a written requirement and materialises a whole runnable project directory on disk. The test for choosing it is whether your goal is a working skeleton from nothing - if so it removes the mechanical work of creating directories, declaring dependencies and writing a README. If you only need to change three lines in an existing file, the prompt overhead outweighs the benefit. This sheet is split into install and create, improve and feedback, and model and flags. Start with --steps, because it decides the internal workflow: clarify asks questions first, tdd writes tests first, and evaluate reviews an implementation without rewriting it, so the flow should follow the nature of the task rather than defaulting blindly. When improving an existing codebase, -i lets you select which files enter the context window, and selecting too much is the most common source of context overflow and drift. For a new project, drop the requirement into a file named prompt inside the project directory and pass only the path.