gpt-coder

About

GPT Coder demonstrates the test-feedback loop that later defined agentic coding, compressed into a single Python script: GPT generates a coding challenge, a solution, and test cases, then the tool compiles, executes, and tests the solution against them, feeding failures back until they pass. The pipeline is fixed rather than open-ended — there is no general tool use, file editing, or interactivity — and configuration is just an OPENAI_API_KEY in a .env file. One commit, no releases, and 26 stars mark it as a ChatGPT-era experiment published in April 2023 and never developed further. Its value now is historical: an early example of test-driven self-refinement that later harnesses industrialized.