Stable Protocol Specification (v1.1.0 Freeze)

This document formalizes the official, frozen architectural layers and design contracts for the MultiModel Dev OS protocol, guaranteeing portability and long-term stability across diverse AI coding agents.

Best for: Third-party client integrations, tool adapter developers, and engineering teams establishing robust AI coding guidelines.

1. Frozen Layers Architecture

MultiModel Dev OS enforces a three-tier design to decouple core workspace context from individual IDE adapters:

┌────────────────────────────────────────────────────────┐
│ LAYER 1: Root Contracts (Single Source of Truth)       │
│ AGENTS.md • MEMORY.md • TASKS.md • RUNBOOK.md          │
│ Status: FROZEN                                         │
└──────────────────────────┬─────────────────────────────┘
                           │ Centralizes project context
┌──────────────────────────▼─────────────────────────────┐
│ LAYER 2: Configuration & Modules (.ai/)                │
│ context/ • agents/ • skills/ • prompts/ • checks/      │
│ Status: FROZEN                                         │
└──────────────────────────┬─────────────────────────────┘
                           │ Routes files dynamically
┌──────────────────────────▼─────────────────────────────┐
│ LAYER 3: Tool & IDE Adapters                           │
│ .cursorrules • CLAUDE.md • .vscode/ • .gemini/        │
│ Status: FROZEN                                         │
└────────────────────────────────────────────────────────┘

2. Layer 1: Root Contracts

The following root files serve as the workspace single source of truth and are completely frozen:

AGENTS.md: Defines team structures, boundaries, and model configurations (Codex project memory / Antigravity AI workflow).
MEMORY.md: Preserves key context decisions, codebase state, and repository milestones.
TASKS.md: Tracks current active goals and backlog.
RUNBOOK.md: Operational scripts and workspace verify procedures.

3. Layer 2: Scaffolding Layout

The following directories and files under .ai/ are strictly locked:

.ai/config.yaml: Standard configuration file mapping templates and active adapters.
.ai/context/: Standard markdown files containing architecture guidelines, briefs, and SEO rules.
.ai/skills/: High-frequency modular instructions for agent workflows.
.ai/session-logs/: Standard directory for tracking agent workspace operations.

4. Layer 3: Adapters Setup

IDE adapters translate centralized rules to tool-specific paths (Cursor project rules / Claude Code project instructions):

Cursor: Linked via root .cursorrules
Claude: Linked via root CLAUDE.md
Gemini: Linked via .gemini/settings.json
VS Code: Linked via .vscode/settings.json

5. Experimental vs. Stable Scope

Stable: Root files, scaffolding directories, CLI subcommand signatures, and standard config structures.
Experimental: Custom check overrides under .ai/checks/ and stack-specific context routing.

Explore our Upgrades & Migration Guide or Release Quality Checklist for staging controls.

Stable Protocol Specification (v1.1.0 Freeze) ​

1. Frozen Layers Architecture ​

2. Layer 1: Root Contracts ​

3. Layer 2: Scaffolding Layout ​

4. Layer 3: Adapters Setup ​

5. Experimental vs. Stable Scope ​