GitHub's Spec Kit passes 100,000 stars: specification first, code second
The open source toolkit billed as the antidote to vibe coding forces AI to produce a structured specification before touching code. Its six-command flow works with more than 30 agents, from Claude Code to Cursor.
GitHub's open source Spec Kit toolkit has gone viral on X again, with posts calling it the fix for vibe coding's biggest problem racking up tens of thousands of views. A fact-check first: the tool is not new. GitHub announced Spec Kit in September 2025; the stars the viral posts say arrived "within days" accumulated over a year and passed 117,000 by July. The real story is the tool itself, because the pain it addresses is real.
The pain: throwing vague prompts at an agent and praying it does not wreck the project. Spec Kit instead imposes spec-driven development: before AI touches code, what is to be built gets written down, and those specifications are kept not as dead documents but as living artifacts the agent returns to at every step.
The flow is six commands: /constitution sets the project's rules and standards, /specify describes what you want to build, /clarify surfaces doubts before starting, /plan lays out the architecture and stack, /tasks breaks the work into ordered tasks, and /implement kicks off execution. The result: less time fixing absurd mistakes, more consistent code, and far more predictable outcomes when working with agents.
The toolkit is not tied to one agent: it works with more than 30 coding agents including Claude Code, GitHub Copilot, Cursor, Codex, Gemini CLI and Windsurf. Read alongside this week's AGENTS.md debate and Uber's software factory, the picture comes together: in the agent era the competition is not in generating code but in describing the work properly before handing it over.
