GitHub has released Spec Kit, an open-source toolkit that introduces spec-driven development (SDD) to AI-assisted coding workflows. Instead of prompting AI agents with vague descriptions, developers write a structured specification first — a 'constitution' capturing non-negotiable project principles — then follow four phases: specify, plan, tasks, and implement. The toolkit provides a CLI, templates, and slash commands that work with GitHub Copilot, Claude Code, and Gemini CLI. Spec Kit maintains persistent project context so multiple developers using different AI tools share a common reference point, reducing inconsistency and AI-generated technical debt. It targets greenfield projects, feature work, and legacy modernization. Critics note it could reintroduce waterfall-like rigidity, but GitHub positions the spec as lean and living rather than a heavyweight requirements document.

5m read timeFrom devops.com
Post cover image
Table of contents
Specs as the Source of TruthHow it WorksWhere it Makes the Most SenseStill Early, But Worth Watching
4.7K Impressions1 Comment