The Problem: "Context Rot" AI-assisted development is magic on Day 1, but by Day 4, LLMs lose track of the broader architecture. Spec-Driven AI addresses this by forcing the AI to track its own state in markdown files, ensuring zero context loss even in large projects.

How it Works: 3-Phase State Management The system implements a strict workflow where every feature moves through three permanent phases:

1. Planning: High-level architectural specs.

2. In-Progress: The active "Context Anchor" for the current session.

3. Executed: A permanent audit trail of why code was written.

✨ Key Features

• Spec-Driven Workflow: Prevents hallucinations by grounding the AI in markdown specs.

• Specialised Skills: Includes the spring-boot-unit-test skill for automatic JUnit 5 generation.

• Auto-Review Agents: java-code-reviewer and api-test-reviewer run automatically to check security and coverage.

• Open Source Core: The methodology and prompt engineering files are MIT-licensed.

🛠 Optional Automation While the workflow is open source, we offer a CLI Toolkit (available for purchase) that automates session management, git summaries, and agent execution.