GSD Core

Git. Ship. Done.

Links

• GitHub: https://github.com/open-gsd/gsd-core
• NPM: https://www.npmjs.com/package/@opengsd/gsd-core
• Docs: https://github.com/open-gsd/gsd-core/tree/main/docs
• Discord: https://discord.gg/mYgfVNfA2r

Overview

GSD Core is a lightweight context-engineering and spec-driven development system for AI coding agents. It is designed to keep the main session lean by pushing research, planning, execution, and verification into fresh-context subagents.

The core idea is a repeatable five-step loop: discuss, plan, execute, verify, and ship. That makes it closer to a disciplined project workflow than a generic prompt pack, with artifacts and phase boundaries intended to reduce context rot on larger tasks.

Key Features

• Multi-runtime support: Claude Code, OpenCode, Gemini CLI, Kimi CLI, Kilo, Codex, Copilot, Cursor, Windsurf, and others
• Fresh-context execution: heavy work runs in subagents with clean context windows
• Phase loop: discuss, plan, execute, verify, ship
• Installer-first setup: npx @opengsd/gsd-core@latest
• Project bootstrap: /gsd-new-project to start a new workflow
• Documentation-driven: tutorials, how-tos, reference docs, and architecture notes

Use Cases

• Orchestrating multi-step coding tasks without letting the main chat get bloated
• Running spec-driven feature work with explicit planning and verification
• Standardizing how teams use different coding agents across runtimes
• Keeping durable project context through structured artifacts instead of ad hoc prompts

Why It Matters

GSD Core targets a common failure mode in agentic coding: once a project grows, context fills up, quality drops, and the agent starts to drift. Its answer is to treat the agent workflow itself as a process that can be structured, repeated, and verified.

That makes it interesting both as a tooling layer and as an example of the broader shift from prompt engineering toward context engineering.