Spec-driven AI coding that actually ships
Stop fighting your AI. Get shit done.
Most AI coding setups rot as the context window fills. GSD keeps the context clean, the work verified, and the git history honest — so you can describe what you want and walk away.
Works on macOS, Windows, and Linux.
Trusted by engineers at
“By far the most powerful addition to my Claude Code. Nothing over-engineered. Literally just gets shit done.”
Two ways to get shit done
Same philosophy, two surfaces. Bring the workflow to the AI tool you already use, or run the standalone agent that owns the whole loop.
get-shit-done
get-shit-done-ccA spec-driven prompting system that drops into the AI tool you already use.
- ›Works across 15+ runtimes — Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Copilot, OpenCode & more
- ›Six commands, one loop: new-project → discuss → plan → execute → verify → ship
- ›Solves context rot by doing heavy work in fresh subagent contexts
- ›Structured artifacts (PROJECT, ROADMAP, STATE, CONTEXT) survive session boundaries
gsd-2
gsd-piA real coding agent built on the Pi SDK — direct control over context, sessions, and execution.
- ›One command — /gsd auto — runs an entire milestone autonomously
- ›Clears context between tasks and injects exactly the right files at dispatch
- ›Manages git branches, atomic commits, cost & token tracking
- ›Detects stuck loops, recovers from crashes, and runs headless in CI
From asking to doing
The original GSD went viral as a prompt framework. It worked — but it was fighting the tool, hoping the LLM would follow instructions. gsd-2 controls the harness itself.
- Injects prompts through slash commands
- Hopes the LLM follows the instructions
- No real control over the context window
- No control over sessions or execution
- Clears context between tasks, deterministically
- Injects exactly the right files at dispatch time
- Manages branches, commits, cost & tokens
- Detects stuck loops and recovers from crashes
One command. Walk away. Come back to a built project with clean git history.
Why it works
Three things most AI-coding setups get wrong — and how GSD fixes each.
No context bloat
As a session grows, quality degrades. GSD keeps your main context clean by doing the heavy work in fresh subagent contexts — researchers, planners, and executors each start with exactly what they need.
Shared memory
Structured artifacts survive session boundaries: vision, scope, roadmap, current position, and per-phase decisions. Every new session loads them and knows exactly where things stand.
Real verification
Code that 'runs' isn't code that 'works'. GSD walks through what was built, diagnoses failures with dedicated debug agents, and generates fix plans before a phase is declared done.
Pick your surface and start
If you know clearly what you want, this will build it for you.