Most failures with AI coding agents aren't the model being dumb — they're the model being starved. Hand it a clever one-line prompt and it guesses; hand it the same structured context a new human hire would get and it ships. This repo is the smallest concrete demonstration of that bet: a folder of conventions plus two slash commands, nothing you install.
What Sets It Apart
- It splits planning from doing.
/generate-prpreads your INITIAL.md and the codebase to write a Product Requirements Prompt — a full implementation blueprint — and only then does/execute-prpbuild against it. So the expensive reasoning happens once, reviewably, before a line of code exists. - PRPs bundle documentation links, known gotchas, and explicit validation gates the agent must pass, so it self-corrects in a loop instead of handing you code that silently breaks.
- The
examples/folder is a first-class input: the agent mimics your existing patterns rather than inventing its own conventions. - It is deliberately framework-free — plain Markdown files (CLAUDE.md, INITIAL.md, PRPs/) any Claude Code user can copy in an afternoon.
Great Fit / Look Elsewhere
Great fit if you already use Claude Code and keep re-explaining the same conventions every session, or you want repeatable multi-step feature builds rather than chat-by-chat coding. Look elsewhere if you want a runtime library, an autonomous agent that manages its own memory, or support beyond Claude Code — this is a methodology and a starter template, not a tool you run.
