Project Overview

The AI coding assistant space has become increasingly crowded, but most tools in this category share a common blind spot: they treat context windows as infinite resources. GSD (Get Shit Done) takes the opposite bet — that context rot is the fundamental bottleneck in AI-assisted development, and that managing it requires explicit, structured workflows rather than hoping the model figures it out. What sets GSD apart from the dozens of ‘AI coding agent’ projects on GitHub is its laser focus on the orchestration layer. Rather than wrapping a specific model or IDE plugin, it provides a meta-prompting and context engineering system that sits on top of existing tools like Claude Code, Copilot, or Cursor. With over 60,000 stars[1], the project has clearly struck a nerve with developers who’ve experienced the frustration of watching an AI assistant’s output quality degrade as conversations grow longer. The architecture is opinionated about one thing above all else: keeping the main context window lean by farming out execution to fresh subagent contexts.

What It’s For

GSD is designed for developers who treat AI as their primary coding engine rather than a copilot — solo builders, indie hackers, and small teams who need consistent quality across long development sessions. If you’ve ever had Claude or Copilot start strong on a feature only to lose coherence halfway through, GSD’s six-command workflow directly addresses that degradation. The system is particularly well-suited to greenfield projects where you can define requirements upfront, though the codebase mapping feature does accommodate existing projects. Where it might fall short is in highly exploratory or experimental work where requirements shift constantly — the structured phase-by-phase approach trades flexibility for reliability. Compared to alternatives like SpecKit or OpenSpec, which target larger teams with sprint ceremonies, GSD deliberately strips away the organizational overhead. There are no story points, no stakeholder syncs, no Jira integrations. The complexity lives in the prompt engineering and subagent orchestration, not in the user’s workflow.

How to Use It

The core workflow is a six-command loop that maps roughly to the software development lifecycle. You start with /gsd-new-project which runs research and generates a requirements document and roadmap — but only after you’ve approved the direction. For existing codebases, /gsd-map-codebase analyzes your stack and conventions first so the project initialization asks relevant questions. Each phase then follows a discuss-plan-execute-verify pattern: /gsd-discuss-phase captures your design decisions before planning begins, /gsd-plan-phase runs a research-plan-verify loop until the plans are solid, and /gsd-execute-phase spins up parallel subagents that each get a fresh 200k-token context. The key architectural insight is that execution happens in isolated contexts — your main session stays at 30-40% capacity while the heavy lifting happens elsewhere. Each task gets its own atomic commit, and verification walks through what was built with automatic fix plans for anything broken.

Initializes GSD in your project — this is the single entry point that sets up the entire workflow system

npx get-shit-done-cc@latest

Runs research and generates a requirements document and roadmap after you approve the project direction

/gsd-new-project

Analyzes existing codebase stack, architecture, and conventions before starting a new project

/gsd-map-codebase

Recent Updates

Latest Release: v1.50.0-canary.2 (Not specified)

Canary release with experimental features — the project is actively iterating on the core workflow

The project has seen rapid iteration with multiple release candidates and canary versions, suggesting active development and community feedback incorporation. The Discord community and $GSD token indicate an ecosystem approach beyond just the CLI tool, which is unusual for a developer tool and worth watching for how it affects project governance.

Sources & Attributions

[1] 60,576 stars on GitHub as of the repository data — gsd-build/get-shit-done [2] Subagent architecture with fresh 200k-token contexts per task — gsd-build/get-shit-done README