get-shit-done: Streamlining AI-Powered Development with Meta-Prompting and Context Engineering

Introduction

get-shit-done is an acclaimed, lightweight, and powerful meta-prompting, context engineering, and spec-driven development system, primarily designed for Claude Code and other LLMs. Developed by TÂCHES, this project aims to solve the common issue of 'context rot,' where AI model quality degrades as context windows grow. With over 63,000 stars, it provides a structured workflow to transform ideas into reliable, high-quality code, trusted by engineers at major tech companies.

Installation

To get started with get-shit-done, simply open your terminal and run the following command:

npx get-shit-done-cc@latest

The installer will guide you through selecting your preferred runtime, for example, Claude Code, OpenCode, or Gemini CLI, and the installation scope, either global or local. For a seamless experience, especially with Claude Code, it is recommended to run your AI with permission skipping enabled:

claude --dangerously-skip-permissions

This allows get-shit-done to operate without constant manual approvals, ensuring a fluid development workflow. You can also find detailed instructions for non-interactive installations in the official README.

Examples

get-shit-done guides you through a structured development lifecycle, from initial idea to deployment. Here’s a glimpse of the core workflow commands:

# 1. Initialize a new project:
/gsd-new-project

# 2. Discuss the current phase to capture implementation decisions:
/gsd-discuss-phase 1

# 3. Plan the phase, including research and task breakdown:
/gsd-plan-phase 1

# 4. Execute the planned tasks in parallel waves:
/gsd-execute-phase 1

# 5. Manually verify the implemented work:
/gsd-verify-work 1

# Automatically advance to the next logical step:
/gsd-progress --next

For ad-hoc tasks that don't require the full planning cycle, the /gsd-quick command offers a faster path with the same quality assurance:

/gsd-quick
> What do you want to do? "Add dark mode toggle to settings"

Why Use

get-shit-done stands out due to several key architectural principles that ensure high-quality, reliable AI-driven development:

• Context Engineering: The system meticulously manages context by generating and utilizing specific markdown files, such as PROJECT.md for vision, REQUIREMENTS.md for scope, and STATE.md for session memory. This ensures that the AI always operates with a fresh, relevant context, preventing quality degradation.
• XML Prompt Formatting: All plans are structured in optimized XML format, providing precise, unambiguous instructions to the AI. This eliminates guesswork and embeds verification steps directly into the tasks.
• Multi-Agent Orchestration: A thin orchestrator coordinates specialized agents for research, planning, execution, and verification. This parallel processing allows complex phases to be completed efficiently, keeping the main context window lean and responsive.
• Atomic Git Commits: Every task completed by the system results in its own atomic Git commit. This provides a clean, traceable history, simplifies debugging with git bisect, and offers clear observability into the AI's automated workflow.
• Modular Design: The system is highly adaptable, allowing developers to add emergency tasks, adjust plans, complete milestones, and start new ones without being locked into a rigid structure.

Links

Explore get-shit-done further through these official resources:

• GitHub Repository
• npm Package
• Join the Discord Community
• Follow on X (Twitter)