GitHub user pangon released an Apache 2.0 licensed repository template designed to structure AI-assisted software development, specifically targeting the pre-coding phases. The scaffold organizes objectives, user stories, requirements, and architecture decisions into a navigable system for Claude Code agents, ensuring clear context for AI agents while maintaining human oversight. It implements a four-phase workflow (Objectives, Design, Code, Deploy) with hierarchical instructions and artifact indexing to minimize context window load. Four core principles include AI-first development, everything-in-repo, context-window efficiency, and decision capture, all aimed at preventing unstructured 'vibecoding' failure modes.

The scaffold starts empty, populating progressively through guided skills like /SDLC-elicit for requirements and /SDLC-design for architecture, with decisions recorded in two-file systems for traceability.