Build multi-phase software projects as a solo dev without losing quality as AI conversations get longer.
Organize complex features into structured phases with clear requirements, roadmaps, and verification steps.
Keep your main AI session lean by offloading heavy work to fresh subagent sessions with clean context.
Maintain a living project snapshot so you can pause, switch tools, and resume without re-explaining everything.
Get Shit Done (GSD) is a workflow system designed to help solo developers build software more reliably with AI coding assistants like Claude Code, Cursor, Windsurf, or Codex. The core problem it solves is "context rot", the well-known issue where the longer a conversation with an AI assistant runs, the worse the quality of its output becomes because the AI's working memory fills up with irrelevant history. GSD addresses this with a structured six-step loop: initialize a project (collecting requirements and generating a roadmap), discuss each phase (capturing design decisions before planning), plan each phase (research, create tasks, verify the plan), execute the phase (run tasks in parallel using fresh AI subagents, each with a clean context), verify the work (walk through what was built and diagnose failures), and then ship (create a pull request). By routing the heavy lifting into dedicated subagent sessions that each start with a clean context window, your main session stays lean and the quality stays consistent. Between sessions, GSD maintains a set of structured planning files, a project vision document, requirements, roadmap, and a current-state snapshot, so that when you start a new session the AI can immediately pick up where it left off without needing you to re-explain the entire project. You would use GSD if you are a solo developer who relies heavily on AI coding tools and finds that complex, multi-phase projects tend to go off the rails as sessions get longer or as context accumulates. It is installed via npm and works as a layer on top of whichever AI coding runtime you prefer. The tech stack is JavaScript (distributed as an npm package), and it works across macOS, Windows, and Linux. No particular backend framework is required, it installs into your existing AI coding workflow.
Generated 2026-05-18 · Model: sonnet-4-6 · Verify against the repo before relying on details.