Spec-Driven Development

Spec-Driven Development is an open-source Claude skill created by FredAntB that solves a common multi-agent coding problem: when Claude Code, Cursor, Copilot, and other AI tools each fill in gaps with their own assumptions, they contradict each other. The skill interviews you before any code is written, generates three structured spec files, and then produces matching configuration files for every AI tool in your stack so they all read from the same source of truth.

What It Is

Spec-Driven Development is a Claude skill — a file you install into Claude.ai or the Claude Code tab — that enforces a spec-first workflow for AI-assisted software development. Rather than letting each AI agent interpret your intent independently, the skill creates a shared "project constitution" that every tool must read before taking action. It is released under the MIT license and hosted on GitHub.

How the Workflow Works

The skill activates on natural-language trigger phrases like "I want to start a new project" or "my AI keeps going off script." Claude then conducts a short four-question interview and generates three files:

• requirements.md — traceable shall-language requirements with REQ-xxx IDs and acceptance criteria
• design.md — architectural decisions and data model
• tasks.md — atomic, ordered implementation steps, each linked back to a requirement ID

For existing codebases, the skill reverse-engineers these files from what you describe, marking inferred fields with [TO VERIFY] and making spec verification Phase 1 before any new code is written.

Cross-AI Config Generation

After the spec files exist, the skill generates a Universal Instruction Block and writes it into tool-specific config files:

• CLAUDE.md for Claude Code
• .cursorrules for Cursor
• .windsurfrules for Windsurf
• .github/copilot-instructions.md for GitHub Copilot
• .aider.conf.yml for Aider

Each config file contains the same mandate: read all three spec files before acting, never implement requirements not in requirements.md, never alter the data model without updating design.md first, and never guess at ambiguous requirements — ask instead. A divergence protocol requires agents to stop and surface conflicts rather than silently invent solutions.

Test Suite and CI

The skill ships with 135 assertions across three test phases, all runnable locally or via GitHub Actions:

• Phase 2A (67 static assertions): automated, hard gate on every push
• Phase 2B (15 behavioral tests): manual live-session evaluation
• Phase 2C (53 generation quality checks): automated against committed fixture files

The repository's CI workflow runs Phase 2A and 2C on every push to main and on every pull request. The README notes that all 135 assertions were written before the corresponding fixes, following a test-first methodology.

Current Status: v1.0-beta

The repository was created in May 2026 and last updated May 26, 2026. The current release is tagged v1.0-beta. The project page describes it as a public beta that has passed all automated assertions and three end-to-end generation flows but has not yet been tested by a broad external audience. The maintainer is actively recruiting five beta testers matching specific profiles: a complete beginner, a solo developer, a team lead using multiple AI tools, a developer with an existing codebase and no specs, and a developer using three or more AI coding tools simultaneously. The repository had 56 stars and 4 forks as of the last recorded update.