Get Shit Done (GSD)

★ 64.8kContext Engineering

#meta-prompting#spec-driven#claude

Overview

A 64.7k+ star meta-prompting, context engineering and spec-driven development system whose motto is 'Git. Ship. Done.' It attacks the most common AI-coding failure mode — context rot, the quality degradation that accumulates as a session's context window fills — by running heavy work in fresh-context subagents. The original gsd-build/get-shit-done repo is archived; development continues as GSD Core in open-gsd/gsd-core, supporting Claude Code, Codex, Cursor, Windsurf and more.

Who it's for

Developers building real projects with coding agents who are frustrated by degrading context and endless revision loops — and who are willing to shift from conversational development to a spec + phase-loop workflow.

What you'll learn

01Core idea: spec first, let Claude get it right once
GSD rests on three pillars: meta-prompting (meta-prompts drive the AI to decompose and plan for itself), context engineering (actively manage context to avoid quality degradation) and spec-driven development (spec first — write down what you want before writing code). The point is not 'write then fix', but nailing down requirements, plans and acceptance criteria before touching code so the coding assistant is close to right the first time. The original repo was described as a lightweight system 'for Claude Code by TÂCHES'.
02What it solves: context rot and the clean-context answer
The pain point it targets is context rot: as a session goes on, the AI's context window fills with history, irrelevant files and intermediate artifacts, silently degrading output quality; meanwhile there is no shared memory between sessions and nothing verifies the code actually works. GSD's answer is to run heavy work in fresh-context subagents — every executor starts with a clean 200k-token context while the main session stays lean, so key reasoning isn't polluted by stale information.
03The five-step phase loop: Discuss → Plan → Execute → Verify → Ship
Every milestone repeats the same loop, one phase at a time: Discuss (capture implementation decisions before anything is planned), Plan (research, decompose and verify the plan fits a fresh context window), Execute (run plans in parallel waves, each executor starting with clean context), Verify (walk through what was built, diagnose and fix before declaring done), then Ship (create the PR, archive the phase, repeat for the next milestone).
04Key artifacts: STATE.md / CONTEXT.md and the verify step
Cross-session memory is carried by structured artifacts: STATE.md records current state and CONTEXT.md preserves decisions and background, surviving session boundaries to fill the 'no shared memory between sessions' gap. The Verify step matters most — it isn't about whether tests ran, but walking through what was built and generating fix plans before a phase is declared done. These three things (clean context, cross-session artifacts, verification) are why the system scales.
05Getting started: installer, commands and tutorials
Getting started is simple: run npx @opengsd/gsd-core@latest. The installer asks which runtime you use (Claude Code, OpenCode, Antigravity CLI, Kimi CLI, Codex, Copilot, Cursor, Windsurf and more) and configures it — the docs explicitly say to use the installer rather than copying files by hand. Then run /gsd-new-project for a greenfield project or /gsd-onboard for an existing codebase, and follow the official tutorial through your first shipped phase.

Practice tips

Install GSD on your primary runtime with the npx @opengsd/gsd-core@latest installer — don't copy files by hand.
Run /gsd-onboard on an existing project and watch how it produces cross-session artifacts like CONTEXT.md and STATE.md.
Walk one milestone through all five phases, paying special attention to Execute's clean context and Verify's walkthrough-and-fix step.
Compare your usual conversational development against GSD's spec-first loop and note the difference in output quality.