Software engineer with 6.5+ years of experience.

leetcode

gumroad

devto

github

Samarth Hathwar  • Peerlist

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:<ol><li>Planning: High-level architectural specs.</li><li>In-Progress: The active "Context Anchor" for the current session.</li><li>Executed: A permanent audit trail of why code was written.</li></ol>✨ Key Features<ul><li>Spec-Driven Workflow: Prevents hallucinations by grounding the AI in markdown specs.</li><li>Specialised Skills: Includes the spring-boot-unit-test skill for automatic JUnit 5 generation.</li><li>Auto-Review Agents: java-code-reviewer and api-test-reviewer run automatically to check security and coverage.</li><li>Open Source Core: The methodology and prompt engineering files are MIT-licensed.</li></ul>🛠 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.