Definition: An open-source, spec-driven development system for AI coding agents, focused on context engineering to keep an agent on track across long or multi-session builds. It sits on top of an existing AI coding tool (such as Claude Code) and adds a repeatable phase loop, specialist subagents, and persistent planning artefacts, rather than replacing the editor or assistant.
So what? Long agentic coding sessions tend to suffer from "context rot", where output quality drops as the context window fills with accumulated discussion; GSD is aimed at managing that so work stays reliable over time.
Example: Running planning and execution steps through fresh-context subagents, with state files that persist across a context reset so the agent does not lose track of requirements partway through a build.
So what? Long agentic coding sessions tend to suffer from "context rot", where output quality drops as the context window fills with accumulated discussion; GSD is aimed at managing that so work stays reliable over time.
Example: Running planning and execution steps through fresh-context subagents, with state files that persist across a context reset so the agent does not lose track of requirements partway through a build.
