gsd-build/get-shit-done

⭐ 61,244  ·  JavaScript  ·  GitHub Repo

A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code by TÂCHES.

claude-code context-engineering meta-prompting spec-driven-development

1-Sentence Summary

GSD kills context rot for solo devs by offloading heavy work to fresh subagent sessions.

🔥 Key Capabilities & USP

• Six-Command Workflow Loop — A repeatable cycle (Initialize → Discuss → Plan → Execute → Verify → Ship) that eliminates decision fatigue. Each command does exactly one thing, so you never wonder "what's next?".
• Parallel Subagent Execution — Plans execute in parallel waves, each task gets its own fresh 200k-token context and atomic commit. This is the core solution to context rot: your main session stays at 30–40% utilization while heavy lifting happens in clean environments.
• Structured Shared Memory — Persistent artifacts (PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md, CONTEXT.md) survive session boundaries. Close your terminal, come back tomorrow, and GSD knows exactly where you left off.
• Built-in Verification & Debug Agents — After execution, GSD walks through what was built, diagnoses failures with dedicated debug agents, and generates fix plans before declaring a phase done. No more "it worked in my session" surprises.
• Multi-Runtime Support — Works with Claude Code, OpenCode, Gemini CLI, Kilo, Codex, Copilot, Cursor, Windsurf, and more via a single installer. Your workflow doesn't change when you switch tools.

USP: Unlike enterprise spec-driven tools that require Jira, sprint ceremonies, and team buy-in, GSD is designed for solo developers and small teams who want a lightweight, frictionless workflow without the overhead.

Technical Architecture

Component	Description
Runtime	JavaScript (Node.js), installed via npx
Orchestration	Multi-agent system with dedicated subagents (researcher, planner, executor, verifier)
Context Engineering	XML prompt formatting, subagent orchestration, state management
Storage	File-based structured memory (PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md, CONTEXT.md)
Configuration	.planning/config.json with settings for mode (interactive/yolo), model profiles (quality/balanced/budget), workflow toggles, and parallelization
Installation	Installs to runtime-specific skill directories (e.g., ~/.claude/skills/gsd-*/)
Git Integration	Each task gets its own atomic commit, maintaining clean git history

Quick Start Guide

# Install and initialize GSD for Claude Code
npx get-shit-done-cc@latest

# Launch Claude Code with full automation (requires --dangerously-skip-permissions)
claude --dangerously-skip-permissions

# Start a new project
/gsd-new-project

# Map your existing codebase (use when returning to a project)
/gsd-map-codebase

# The six-command workflow loop
/gsd-discuss-phase 1     # Discuss what needs to be built
/gsd-plan-phase 1        # Plan the implementation
/gsd-execute-phase 1     # Execute the plan in parallel subagents
/gsd-verify-work 1       # Verify the work was done correctly
/gsd-ship 1              # Ship the completed phase

# Milestone management
/gsd-complete-milestone  # Mark current milestone as done
/gsd-new-milestone       # Start a new milestone
/gsd-progress --next     # See what's coming next

# Global installation (custom config directory)
CLAUDE_CONFIG_DIR=/home/youruser/.claude npx get-shit-done-cc --global

Pros, Cons & Use Cases

Pros

• Solves context rot — The single biggest pain point in long AI coding sessions, addressed elegantly through subagent orchestration
• Lightweight — Six commands, no Jira, no sprint ceremonies, no team meetings
• Cross-runtime — Works with Claude Code, Copilot, Cursor, Windsurf, and more
• Frictionless automation — --dangerously-skip-permissions mode enables fully autonomous execution
• Clean git history — Atomic commits per task make rollbacks and code review trivial

Cons

• Requires --dangerously-skip-permissions for full automation — a security tradeoff that may not be acceptable in all environments
• Context rebuilding overhead — Returning to a project requires running /gsd-map-codebase then /gsd-new-project to re-establish state
• Solo/small-team focus — Not designed for enterprise orgs with complex access controls, compliance requirements, or large distributed teams

Who should NOT use this?

• Enterprise teams with strict compliance requirements, multi-role access controls, or existing project management toolchains (Jira, Asana, Linear)
• Developers who never use AI coding assistants — GSD is purpose-built for AI-assisted development workflows
• Teams that need real-time collaboration — GSD is designed for individual sessions, not concurrent multi-user editing
• Projects requiring manual code review gates — The automation model assumes trust in the AI's output

Ideal Use Cases

• Solo developers building side projects, MVPs, or personal tools with AI assistance
• Small teams (2–5 people) using Claude Code or Copilot who want a repeatable workflow without ceremony
• Rapid prototyping — The "Discuss → Plan → Execute → Verify → Ship" loop is perfect for iterating quickly
• Open source maintainers who want to automate routine development tasks while maintaining clean commit history
• AI-assisted codebase refactoring — Fresh subagent contexts prevent context poisoning during large-scale changes

Community & Activity

With 61,244 stars, GSD has clearly struck a nerve in the developer community. This isn't just a niche tool — it's a movement. The project is actively maintained (last updated May 2026), and the rapid adoption signals that context rot is a pain point many developers have been feeling but couldn't articulate. The cross-runtime support means it's not tied to any single AI tool's fate, and the active development suggests new features and runtime integrations are on the horizon. If you're building with AI and haven't tried GSD yet, you're leaving productivity on the table.