GSD Core
A spec-driven development framework for coding agents: a discuss, plan, execute, verify and ship loop run in fresh contexts
Install
npx @opengsd/gsd-core@latestThe installer asks for the runtime and install scope.
This is third-party code. Review the repository files before installing.
What it does
GSD Core installs a set of /gsd-* commands, skills, subagents and hooks that drive a project phase by phase. Each phase has five steps: decisions are captured, a plan is drafted and checked, tasks run in parallel waves in fresh-context subagents, the result is verified with fix plans, then a PR is created. Heavy work goes to subagents so the main session stays lean, and state between sessions lives in files like STATE.md and CONTEXT.md. There are flows for greenfield projects and for onboarding an existing codebase.
Who it is for. For developers running large tasks with an agent who want a disciplined process instead of long chats.
Good fit when
- A multi-phase project where the agent degrades by the end of a session
- You need to onboard an agent to an existing repo with a code map
- You want verification before a PR is created
- Work spans many sessions and state must persist
Not a fit when
- A small fix of a few lines
- You do not want an installer adding hooks and files to your agent config
Example request
/gsd-onboard, then plan the first phase: OAuth sign-inLimitations
Requires Node.js 18+ and npm; the authors advise against copying files by hand because the installer converts them per runtime. Parallel subagents use noticeably more tokens than a single session. The installer registers Claude Code hooks, including an update check at session start.
How to disable. Remove the gsd commands, skills, agents and hooks from ~/.claude or the project's .claude folder, then remove the hook entries from settings.json.
Security check
- Installs hooks that run on agent session events
- Subagents write code, commit and open PRs
README in short
The README explains that GSD Core fights quality loss as the context fills up by moving research, planning and execution into subagents. It describes the five-step phase loop and a single npx install with runtime and scope selection. Docs are split into tutorials, how-to recipes, command and config reference, and architecture explanations. MIT licensed.
FAQ
Which agents are supported?
Per the README: Claude Code, OpenCode, Antigravity CLI, Kimi CLI, Kilo, Codex, Copilot, Cursor, Windsurf and more. The installer asks which one to target.
Can I install both globally and per project?
Yes, but on name clashes the global skill wins the command. The installer and /gsd-health warn about it.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Anthropic's coding agent for the terminal, IDE and GitHub: understands your codebase, runs tasks and handles git