Feeding AI the screenplay instead of a sticky note

What it does This repository is a template for “context engineering” — the practice of giving AI coding assistants comprehensive background material rather than terse prompts. It provides scaffolding for project-wide rules (CLAUDE.md), code examples, and feature requests (INITIAL.md) that get compiled into detailed Product Requirements Prompts (PRPs). These PRPs act as implementation blueprints with validation gates, test requirements, and error handling patterns meant to be consumed by Claude Code.

The interesting bit The core argument is that most AI agent failures are context failures, not model failures. The template forces you to write a full screenplay — documenting conventions, linking APIs, and seeding examples — so the AI has enough information to handle multi-step features and self-correct through test loops. It is essentially onboarding documentation for a new team member who happens to be an LLM.

Key highlights

• Includes CLAUDE.md for global rules and an examples/ folder to teach the AI your project’s patterns
• PRPs include implementation steps, validation gates, and a confidence score (1–10) before execution
• Custom Claude Code commands in .claude/commands/ automate the research and implementation loop
• Provides an example multi-agent PRP and templates for structuring feature requests
• Explicitly designed to improve consistency and reduce “context failures” by front-loading documentation

Caveats

• The template is tightly coupled to Claude Code (.claude/ directory, slash commands); portability to other assistants is theoretical
• RAG and external tool integrations are explicitly excluded for now
• The “10x better than prompt engineering” claim is a qualitative opinion, not a benchmark

Verdict Grab this if you are tired of AI assistants ignoring your project conventions or hallucinating file structures. If you do not use Claude Code or prefer minimal prompts, the strategy is sound but the tooling will not fit.