AI and MCP guidance

Generate intent, then lower it

AI should normally describe the user's goal in high-level terms first. A compiler or approved lowering step turns that intent into Atomic Protocol V2.

Current AI rules

  • Prefer explicit V2 fields over guessed shortcuts.
  • Ask for missing composition dimensions, timing, asset references, and text font information.
  • Use anchor for a pivot and translate.z for world position.
  • Use zIndex for paint order and do not treat it as world distance.
  • Keep fill opacity and stroke opacity independent.
  • Keep effect order explicit.
  • Do not invent renderer, shader, texture, or buffer fields.
  • Keep future high-level features labeled as intent until they have lowering rules.

A good patch shape

{
  "operation": "update",
  "target": "layers.hero.transform",
  "set": {
    "anchor": { "x": 0.5, "y": 0.5, "z": 0 },
    "rotate": { "x": 0, "y": 0, "z": -4 }
  }
}

The example is a teaching block. The editor or MCP contract decides how a patch is validated and applied.

Planned vocabulary

Patterns, presets, templates, and components may be useful high-level concepts. They are not current atoms. Do not emit them inside an atomic layer payload until Compiler A has an approved lowering rule.