open-gsd/gsd-core

GSD Core helps you get better results from AI coding assistants like Claude Code, Cursor, Copilot, and others. The core problem it solves is that when you give an AI assistant too much to work on in a single session, the quality of its output quietly degrades, it forgets earlier instructions, loses track of decisions, and starts producing worse code. This project keeps everything organized and disciplined so that doesn't happen. It works by breaking your work into milestones, each going through the same five-step loop: discuss, plan, execute, verify, and ship. You first capture decisions, then research and plan the work. The actual coding happens in fresh "subagent" sessions that each start with a clean slate, so there's no accumulated baggage. After execution, there's a verification step where the AI walks through what it built and fixes problems before anything is declared done. Finally, it creates a pull request and moves on to the next milestone. Key decisions and context are saved in structured files that persist across sessions, so nothing gets lost when you start fresh. This is for developers, founders, or anyone using AI coding assistants who has hit a wall where the AI's output gets worse as projects grow. If you've ever had an AI assistant build something that doesn't actually work, or if you've lost track of decisions across multiple sessions, this is designed to address those pain points. It's especially relevant for larger projects where you need real structure rather than just firing off prompts. The project is notably runtime-agnostic, it works across many different AI coding tools rather than being locked into one. It's installed via a single command and supports global or local setup. The README emphasizes that the installer is required rather than manually copying files, which ensures cross-runtime compatibility. The entire system is open source under the MIT license and has an active community around it.