MultiModel Dev OS

The Problem

You use Cursor for autocomplete, Claude Code for terminal ops, Gemini for deep audits. Every tool switch loses your context. Every .cursorrules / CLAUDE.md / .vscode/settings.json change drifts out of sync with the others. Prompts bloat, tokens waste, onboarding breaks.

The Fix: 30 Seconds

npx multimodel-dev-os@latest init

Your workspace now has a single source of truth that every AI coding tool reads automatically — no config duplication, no context loss, no vendor lock-in.

Already have a project? Onboard it safely:

npx multimodel-dev-os@latest onboard analyze

Packages

The primary public package is published on npm:

npm install -g multimodel-dev-os

An optional GitHub Packages mirror is maintained under the scoped name @rizvee/multimodel-dev-os. It uses https://npm.pkg.github.com; visibility and access are controlled by GitHub Packages settings.

The root npm package name remains multimodel-dev-os.

Key Features

| | Feature | What It Does | |:---|:---|:---| | 🔄 | Universal Adapter Sync | Write rules once → auto-syncs to .cursorrules, CLAUDE.md, .vscode/, .gemini/, and more | | ⚡ | Caveman Mode | Slash prompt token overhead by ~79% for tight API budgets | | 🧠 | Intelligence Engine | Hash-compressed memory, feedback learning, self-improvement proposals with HITL safety gates | | 📁 | Repo Onboarding | Analyze existing projects, recommend templates, and bootstrap configs without breaking anything | | 🔧 | Zero Dependencies | Pure Node.js CLI — no runtime, no build step, no package manager lock-in | | 🛡️ | 300+ Quality Gates | Built-in validate, doctor, verify, and Skill OS registry checks with strict structural assertions |

Supported Tools & Agents

| Tool / Agent | Adapter File | Status | |:---|:---|:---| | Codex (OpenAI) | adapters/codex/AGENTS.md | ✅ Full support | | Antigravity (Google DeepMind) | .gemini/settings.json | ✅ Full support | | Cursor | .cursorrules | ✅ Full support | | Claude Code (Anthropic) | CLAUDE.md | ✅ Full support | | Gemini (Google) | GEMINI.md | ✅ Full support | | VS Code (Copilot) | .vscode/settings.json | ✅ Full support | | Cline / Continue / Roo Code | Via adapter registry | 🔌 Adapter-ready | | Aider / Windsurf | Via adapter registry | 🔌 Adapter-ready | | MCP Tools (gcloud, Chrome DevTools) | Via tool registry | 🔌 Registry-ready |

Zero lock-in. Switch tools freely — your context, rules, and memory travel with you.

How It Works

┌─────────────────────────────────────────────────────────────┐
│  LAYER 1: Central Root Contracts (Single Source of Truth)   │
│  AGENTS.md  •  MEMORY.md  •  TASKS.md  •  RUNBOOK.md      │
└─────────────────────────┬───────────────────────────────────┘
                          │
┌─────────────────────────▼───────────────────────────────────┐
│  LAYER 2: Configuration & Intelligence (.ai/)                │
│  context/  agents/  skills/  prompts/  checks/  session/    │
└─────────────────────────┬───────────────────────────────────┘
                          │
┌─────────────────────────▼───────────────────────────────────┐
│  LAYER 3: Engine Workflows & Safety Gates                   │
│  onboard analyze  •  adapter sync  •  improve apply         │
└─────────────────────────┬───────────────────────────────────┘
                          │
┌─────────────────────────▼───────────────────────────────────┐
│  LAYER 4: Tool & IDE Adapters                               │
│  .cursorrules  •  CLAUDE.md  •  .vscode/  •  .gemini/      │
└─────────────────────────────────────────────────────────────┘

Essential Commands

# Initialize & Onboard
npx multimodel-dev-os@latest init --template nextjs-saas
npx multimodel-dev-os@latest onboard analyze

# Scan, Status & Memory
npx multimodel-dev-os@latest scan
npx multimodel-dev-os@latest status
npx multimodel-dev-os@latest memory build

# Sync IDE Adapters
npx multimodel-dev-os@latest adapter sync all --approved

# Run Workflows & Handoffs
npx multimodel-dev-os@latest workflow run repo-health
npx multimodel-dev-os@latest handoff build

# Inspect Skill OS registries (read-only)
npx multimodel-dev-os@latest skill-os status
npx multimodel-dev-os@latest skill-os validate
npx multimodel-dev-os@latest skill-os list skills
npx multimodel-dev-os@latest skill-os list prompts
npx multimodel-dev-os@latest workflow show release-check
npx multimodel-dev-os@latest workflow show operator-weekly-review

📖 Full CLI Reference →

Why Not Just a Manual AGENTS.md?

| Capability | Manual Rules File | MultiModel Dev OS | |:---|:---|:---| | Tool Sync | Manual copy-paste across tools | ✅ Automated dynamic adapters | | Context Budgets | Bloats prompts, wastes tokens | ✅ Caveman Mode cuts ~79% overhead | | Standards | Easy to drift and corrupt | ✅ CLI validate + doctor + 300+ check verify | | Templates | Start from scratch | ✅ 6 production-ready real-world templates | | Model Registry | Hardcoded model names | ✅ Dynamic capability-scored routing presets | | Self-Improvement | None | ✅ Feedback → Proposals → Apply with safety gates | | Onboarding | Manual setup every time | ✅ onboard analyze bootstraps existing repos |

What's New in v4.1

• Skill OS foundation - Declarative metadata for structured prompts, skill registries, tool permission classes, advisory guardrails, workflow references, and draft-only business operator templates.
• RACE+ prompt templates - Reusable prompt definitions for Role, Action, Context, Expectation, Constraints, Output format, Verification, and Next action.
• Read-only Skill OS CLI - Local inspection and validation through skill-os status, skill-os validate, skill-os list, and skill-os show.
• Workflow metadata integration - Optional workflow skill_os references connect workflows to skills, prompts, permissions, guardrails, and required context without changing workflow execution.
• Validation-first safety model - Skill OS metadata is declarative only; it does not execute automation, enforce runtime permissions, or make advisory guardrails block live commands.

Patch note: v4.0.1 updates npm package-page documentation after the v4.0.0 publication. It does not change runtime behavior or CLI behavior.

Full Changelog →

Skill OS Foundation

Skill OS adds a structured, validation-only metadata layer for reusable prompts, skills, permission classes, advisory guardrails, workflow references, and draft-only business operator templates.

The v4.1.0 scope is declarative and local-only:

• RACE+ prompt templates
• Skill registry metadata
• Tool permission metadata
• Advisory guardrail metadata
• Workflow skill_os references
• Read-only skill-os CLI inspection
• Draft-only business operator templates

Skill OS metadata does not execute automation, enforce permissions at runtime, call external tools, send messages, publish content, or change workflow behavior.

Start here: Skill OS CLI · Structured Prompts · Skill Registry · Tool Permissions · Hooks and Guardrails · Business Operator Layer · Migration Guide · Adoption Checklist · Authoring Reference

Gateway Foundation

v4.2.0 prepares the Gateway Foundation release. It is not npm-published yet; v4.1.0 remains npm latest until the maintainer completes the manual publication step.

The release adds deterministic gateway protocol contracts, runtime-readable provider/model registry snapshots, explainable dry-run route planning, resilience simulation, a localhost-only mock gateway runtime, preview-only client configuration plans, and bounded local observability.

The mock gateway can start a local HTTP server, answer /health, list mock models, and serve mock chat/streaming responses. External providers remain metadata-only: MultiModel Dev OS does not currently call model providers, load provider credential values, execute real provider requests, perform retries, perform provider failover, probe local engines, persist circuit-breaker state, or enforce Skill OS permissions at runtime.

Client integration plans remain preview-only. Observability remains local, bounded, redacted, and in-memory; it does not upload telemetry, persist logs, retain prompts/completions by default, probe external providers, or enable live fallback.

Gateway docs: Gateway Architecture | Gateway Protocol | Gateway OpenAI Compatibility | Gateway Runtime Registry | Gateway Routing Engine | Gateway Routing Strategies | Gateway Route Explanations | Gateway Resilience | Gateway Retry Policy | Gateway Circuit Breaker | Gateway Resilience Simulation | Gateway Runtime | Gateway Local Server | Gateway Mock Provider | Gateway Streaming | Gateway Authentication | Gateway Observability | Gateway Usage Accounting | Gateway Cost Estimation | Gateway Request Tracing | Gateway Provider Health | Gateway Audit Events | Gateway Client Integrations | Gateway Client Compatibility | Gateway Client Compatibility Matrix | Gateway Client Configuration | Provider Adapter Contract | Gateway API Reference | Gateway Security Model | v4.2 Known Limitations | v4.2 Release Readiness | v4.2 Gateway Planning

Roadmap

| Version | Focus | Status | |:---|:---|:---| | v2.0.0 | Template Galaxy, Model Registry, Stable Protocol | ✅ Released | | v2.2.0 | Codebase Scanner & Hash-Compressed Memory Engine | ✅ Released | | v2.3.0 | Feedback Learning & Proposal Engine | ✅ Released | | v2.4.0 | Approved Proposal Application Engine | ✅ Released | | v2.5.0 | Repository Intelligence Command Center | ✅ Released | | v2.6.0 | Real-Repo Onboarding & Adapter Sync | ✅ Released | | v2.7.0 | Website, Demo & Distribution System | ✅ Released | | v2.8.0 / v2.8.1 | Interactive TUI Dashboard & Plugin Hooks | ✅ Released | | v2.9.0 | Local Workflow Marketplace & Plugin Catalog | ✅ Released | | v3.0.0 | Trusted Remote Catalog & Registry Governance Layer | ✅ Released | | v3.0.1 | Registry UX & Policy Safety Patch | ✅ Released | | v3.0.2 | Registry Sync Security Hotfix | ✅ Released | | v3.1.0 | Modular Source Layout + Formal Unit Tests | ✅ Released | | v3.2.0 | Stable Modular Build + Package Governance | ✅ Released | | v3.5.0 | Trusted Registry Signing + Provenance Foundation | ✅ Released | | v4.0.0 | Modular CLI, verification engine, registry trust, handler tests, docs/DX hardening | ✅ Released | | v4.1.0 | Skill OS foundation: RACE+ prompts, skill registries, permissions, guardrails, workflow metadata, and read-only inspection | ✅ Released | | v4.2.0 | Gateway Foundation: local mock gateway runtime, provider/model registries, deterministic routing, resilience simulation, client previews, observability, and security hardening | Prepared, npm publish pending |

Full Roadmap →

Future roadmap: Release state · AI OS roadmap · v4.1 Skill OS plan · v4.2 Gateway Planning · Skill OS CLI · Hooks and Guardrails · Business Operator Layer · Skill OS Migration Guide

Documentation & Resources

| Resource | Link | |:---|:---| | 📖 Documentation Portal | rizvee.github.io/multimodel-dev-os | | 🐙 GitHub Repository | github.com/rizvee/multimodel-dev-os | | 📦 NPM Registry | npmjs.com/package/multimodel-dev-os | | 🤖 AI Discoverability | llms.txt | | 🚀 Quick Start | Quickstart Guide | | 🏗️ Architecture | Architecture Deep Dive | | ⚔️ Comparison | vs Alternatives | | 🛡️ Stable Protocol | Protocol Specification |

Contributing & Community

We welcome contributions! Propose new adapters, request templates, improve docs, or report issues.

• 📖 Contributing Guidelines
• 🐛 Report a Bug
• 💡 Request a Feature
• ⭐ Star us on GitHub — it helps others discover this project

License

MIT License. Copyright (c) 2026-present MultiModel Dev OS team.