This repository contains the definitive guide for designing, architecting, and implementing production-grade AI Agent Skills.
AI Agent Skills are packaged sets of instructions, prompt fragments, and deterministic backend code designed to be consumed by an AI coding assistant (like Claude Code, Cursor, Aider, GitHub Copilot). They grant host agents new capabilities, standardized workflows, and integrations with external systems.
- REPORT.md: The comprehensive blueprint and architectural guide for developing AI Agent Skills. It covers:
- Repository structures for AI skills
- High-level architecture pipelines
- Prompt engineering strategies (e.g. build systems, variable injection)
- Execution flows and orchestration using parallel subagents
- Context management to avoid token bloat
- Checklists and lessons learned
Read the REPORT.md for a complete walkthrough of how to architect your own AI Agent Skills, complete with templates and design principles.