AI code helpers operate under hard constraints: finite context windows, sensitivity to ambiguity, and high performance on pattern recognition. Traditional software architecture was designed for human cognitive limits — modularity, readability, naming conventions. ARIA is designed for *AI cognitive limits*, which are fundamentally different.
Instructions file GitHub Copilot
Copilot instructions for toutaio/aria, covering aria framework — github copilot instructions, layers — assign one to every aru, naming — semantic address, manifest — required for every aru and composition patterns — declare every connection.
Instructions file
Claude Code instructions for toutaio/aria, covering aria framework — claude project instructions, 1. abstraction layers — where code lives, 2. atomic responsibility units — what each piece is, 3. naming conventions — semantic address and 4. context manifests — every aru needs one.