"Stop throwing request after request at your agent like crazy, and write the spec first."
That's a line posted on X by data engineer José Siles. What he introduced is spec-kit, an open-source toolkit released by GitHub (github.com/github/spec-kit).
Spec-Driven Development in 5 Commands
The core of spec-kit is five commands, one per step.
| Command | Role |
|---|---|
/speckit.constitution | Define the project's principles |
/speckit.specify | Specify what to build |
/speckit.plan | Design the tech stack and architecture |
/speckit.tasks | Break the work into an actual task list |
/speckit.implement | Have the agent carry out the implementation |
Rather than having an agent spit out code first and a person fix it afterward, it enforces the order of spec, plan, task breakdown, implementation. Siles introduced it with the line "DESCANSE EN PAZ VIBE CODING" (rest in peace, vibe coding) (original post, published at 5:02 a.m. KST on August 14, 2026).
How Widely Is It Actually Used?
Checking directly through the GitHub API, the github/spec-kit repository had 127,958 stars as of August 14. That total built up in under a year after it was created on August 21, 2025. It also matches the "more than 125,000" Siles mentioned.
The repo's official description is "Toolkit to help you get started with Spec-Driven Development." It is said to be compatible with more than 30 coding agents, including Copilot, Claude, Codex, and Cursor.
Why This Workflow Now?
"Vibe coding," where you keep making open-ended requests to an agent, is fast, but consistency tends to fall apart as a project grows. spec-kit approaches this problem by first creating an agreed-upon document (the spec) and having the agent work from that document.
For organizations that use AI coding agents at the team level, the habit of documenting agreement on what to build before the code could be a place to cut collaboration costs.
Measure Spec Kit's Impact Yourself
Siles, who introduced the tool, is a personal content account, not an official GitHub spokesperson. Still, key facts such as the five-command setup, the list of compatible agents, and the star count were verified directly against GitHub's official repository.
Quantitative data on the actual benefits of adoption (development speed, bug reduction, and so on) has not been officially published. Each team has to judge the workflow's effectiveness through its own pilot.