GSD (Get Shit Done)

Orchestrate AI agents to manage spec-driven software development, from planning and research to code execution and verification, directly from your command line.

Orchestrate AI agents to manage spec-driven software development, from planning and research to code execution and verification, directly from your command line.

The gist

GSD (Get Shit Done) is an open-source meta-prompting and spec-driven development system. Built by a solo developer, it solves context degradation in large language models by orchestrating specialized AI agents to handle the entire software development lifecycle. It structures the process of building with AI, from high-level goals to atomic, verifiable tasks, without the boilerplate of enterprise software methodologies. It's designed for developers who want to build consistently with tools like Claude Code.

What it does

  • Initializes new projects by turning ideas into structured roadmaps, requirements, and research artifacts.
  • Orchestrates specialized AI agents for planning, research, code execution, and verification.
  • Breaks development phases into atomic plans, each executed in a fresh context window to maintain quality.
  • Manages codebase analysis, allowing it to understand existing projects before adding new features.
  • Executes tasks in parallel waves based on dependencies, with each task getting an atomic Git commit.
  • Provides commands for quick ad-hoc tasks, experimental spikes, and UI sketching.

How it works

GSD is a command-line tool installed via npx. Users interact with it through a series of commands (e.g., /gsd-new-project, /gsd-plan-phase). The system uses these inputs to spawn and coordinate AI agents that generate planning documents, write code, and run verifications. It manages state through a series of markdown and JSON files in a local .planning/ directory. The tool is open-source and works with various code-generating models like Claude Code and Gemini.

Best for

This tool is ideal for solo developers and small teams who want to use large language models for spec-driven development without the overhead of complex enterprise methodologies.

Watch out for

The intended workflow encourages running the underlying LLM's CLI with permissions skipped for automation, which may not be suitable for all security-conscious environments.