English | 简体中文

newtype-profile

AI Agent Collaboration System for Content Creation

Based on oh-my-opencode, redesigned for content creation scenarios.

---

Created by huangyihe (黄益贺)

• YouTube: https://www.youtube.com/@huanyihe777
• Twitter: https://x.com/huangyihe
• Substack: https://newtype.pro/
• 知识星球: https://t.zsxq.com/19IaNz5wK

---

Overview

newtype-profile is an AI Agent collaboration framework designed for content creation. Unlike oh-my-opencode which focuses on code programming, this project redefines the Agent system as an editorial team model, suitable for:

• 📚 Knowledge base management
• ✍️ Article writing and editing
• 🔍 Information research and fact-checking
• 📄 Document extraction and organization

Agent Team

Agent	Role	Default Model	Description
chief	Editor-in-Chief	Claude Opus 4.5 Thinking High	Dual-mode: exploration partner + task coordinator
deputy	Deputy Editor	Claude Sonnet 4.5	Executes specific delegated tasks
researcher	Intelligence Officer	Gemini 3 Pro High	Broad search, discover new information
fact-checker	Verifier	Gemini 3 Pro High	Validate sources, assess credibility
archivist	Librarian	Claude Sonnet 4.5	Knowledge base retrieval, find connections
extractor	Formatter	Gemini 3 Flash	PDF/image/document extraction
writer	Writer	Gemini 3 Pro High	Content production, article drafting
editor	Editor	Claude Sonnet 4.5	Content refinement, structure optimization

Quick Start

Prerequisites

1. Install OpenCode
2. Install Bun (only needed for local development)

Installation

Method 1: npm Package (Recommended)

Step 1: Install the package:

cd ~/.config/opencode
bun add newtype-profile

Step 2: Edit ~/.config/opencode/opencode.json:

{
  "plugin": [
    "newtype-profile"
  ]
}

To update to the latest version:

cd ~/.config/opencode
bun update newtype-profile

Method 2: Clone to Local (Development)

For development or customization:

git clone https://github.com/newtype-01/newtype-profile.git
cd newtype-profile
bun install
bun run build

Then reference the local path in your config:

{
  "plugin": [
    "/path/to/newtype-profile"
  ]
}

Configure Agent Models

Create or edit the newtype-profile config file.

User-level: ~/.config/opencode/newtype-profile.json

Project-level: <project>/.opencode/newtype-profile.json

{
  "google_auth": true,
  "agents": {
    "chief": { "model": "google/antigravity-claude-opus-4-5-thinking-high" },
    "researcher": { "model": "google/antigravity-gemini-3-pro-high" },
    "fact-checker": { "model": "google/antigravity-gemini-3-pro-high" },
    "archivist": { "model": "google/antigravity-claude-sonnet-4-5" },
    "extractor": { "model": "google/antigravity-gemini-3-flash" },
    "writer": { "model": "google/antigravity-gemini-3-pro-high" },
    "editor": { "model": "google/antigravity-claude-sonnet-4-5" }
  }
}

Authenticate Google Antigravity

opencode auth login
# Select Provider: Google
# Select Login method: OAuth with Google (Antigravity)

Usage

Start OpenCode

opencode

Three-Layer Architecture

User ↔ Chief (Editor-in-Chief)
           ↓ chief_task
       Deputy (Deputy Editor)
           ↓ chief_task
       Specialist Agents (researcher, writer, editor...)

You only interact with Chief. Chief automatically coordinates the team:

• Mode 1 - Thought Partner: When exploring ideas, Chief thinks WITH you, challenges flawed logic, and sparring.
• Mode 2 - Execution Coordinator: When you have clear deliverables, Chief decomposes, delegates, and delivers.

Example Conversations

# Research request - Chief delegates to researcher via Deputy
"Help me understand the AI development trends in 2024"

# Writing request - Chief coordinates writer → editor pipeline
"Write an article about this topic based on our research"

# Fact-checking request - Chief dispatches fact-checker
"Verify the sources in this document"

# Complex task - Chief orchestrates multiple agents
"Create a comprehensive report on [topic] with verified sources"

Task Categories

Chief uses chief_task to delegate tasks by category:

Category	Purpose	Model Configuration
research	Information research, trend discovery	Gemini 3 Pro High, temp 0.5
fact-check	Source verification, credibility assessment	Gemini 3 Pro High, temp 0.2
archive	Knowledge base retrieval, document linking	Claude Sonnet 4.5, temp 0.3
writing	Content creation, article drafting	Gemini 3 Pro High, temp 0.7
editing	Content refinement, structure optimization	Claude Sonnet 4.5, temp 0.3
extraction	PDF/image content extraction	Gemini 3 Flash, temp 0.2
quick	Simple quick tasks	Gemini 3 Flash, temp 0.3

Configuration

Model Selection

All models are accessed via Google Antigravity. Available models:

Gemini Series

• google/antigravity-gemini-3-pro-high - High quota Pro version
• google/antigravity-gemini-3-pro-low - Low quota Pro version
• google/antigravity-gemini-3-flash - Fast response version

Claude Series (via Antigravity)

• google/antigravity-claude-opus-4-5-thinking-high - High thinking budget Opus
• google/antigravity-claude-opus-4-5-thinking-medium - Medium thinking budget Opus
• google/antigravity-claude-opus-4-5-thinking-low - Low thinking budget Opus
• google/antigravity-claude-sonnet-4-5 - Sonnet 4.5
• google/antigravity-claude-sonnet-4-5-thinking-high - High thinking budget Sonnet

Custom Agent Settings

Override default settings in your config file:

{
  "agents": {
    "writer": {
      "model": "google/antigravity-claude-sonnet-4-5",
      "temperature": 0.8,
      "prompt_append": "Please use a concise and lively writing style"
    }
  }
}

Disable Specific Agents

{
  "disabled_agents": ["fact-checker", "extractor"]
}

Disable Specific Hooks

{
  "disabled_hooks": ["comment-checker", "agent-usage-reminder"]
}

MCP Server Configuration

The plugin includes built-in MCP (Model Context Protocol) servers. Configure them in your newtype-profile.json:

{
  "mcp": {
    "tavily": {
      "api_key": "tvly-your-api-key"
    },
    "firecrawl": {
      "api_key": "fc-your-api-key"
    },
    "filesystem": {
      "directories": ["~/Documents", "~/Projects"]
    },
    "sequential-thinking": true
  }
}

MCP Server	Default	Required Config	Description
websearch (Exa)	Enabled	None	Web search via Exa.ai
sequential-thinking	Enabled	None	Structured problem-solving
tavily	Disabled	api_key	Advanced web search, crawl, extract
firecrawl	Disabled	api_key	Web scraping and content extraction
filesystem	Disabled	directories	Local file system access

Get API keys:

• Tavily: tavily.com
• Firecrawl: firecrawl.dev

To disable a built-in MCP:

{
  "disabled_mcps": ["sequential-thinking"]
}

Built-in Skills

The plugin includes specialized skills that can be invoked via /skill <name> or /<skill-name>:

Skill	Command	Description
playwright	/playwright	Browser automation via Playwright MCP - web scraping, testing, screenshots
super-analyst	/super-analyst	Elite analytical consulting system with 12 professional frameworks
super-writer	/super-writer	Professional content creation with 6 writing methodologies

Super Analyst Features:

• 7-stage systematic workflow (Problem Understanding → Intelligence Planning → Gathering → Framework Selection → Analysis → Output)
• 12 professional analysis frameworks with detailed prompts
• Automatic complexity detection (Level 1/2/3)
• Bilingual search strategy (Chinese/English)
• Sequential Thinking integration for deep reasoning

Super Writer Features:

• 3-phase streamlined workflow (UNDERSTAND → PREPARE → CREATE)
• 6 professional writing methodologies:
  • W.R.I.T.E - World-building, Relevance, Information, Takeaway, Engagement
  • AIDA Model - Attention, Interest, Desire, Action
  • Content Writing Process - Research-driven structured content
  • Content Creation Techniques - Hook, story, offer framework
  • High-Value Content Strategies - Authority-building content
  • Storytelling Framework - Narrative-driven engagement
• Optional style mimicking (only when user provides reference)
• Automatic complexity detection (simple/complex)
• Quality checklists for each methodology

To disable a built-in skill:

{
  "disabled_skills": ["super-analyst", "super-writer"]
}

Features Inherited from oh-my-opencode

This project retains core capabilities from oh-my-opencode:

• ✅ Background Tasks: Run multiple agents in parallel
• ✅ Todo Enforcement: Ensure task completion
• ✅ Session Recovery: Automatic error recovery
• ✅ Claude Code Compatibility Layer: Support for hooks, skills, commands
• ✅ LSP Tools: Code navigation and refactoring
• ✅ AST-Grep: Code pattern search
• ✅ MCP Support: Extended capabilities

Switch Between Plugins

Use the /switch command to switch between OpenCode plugins:

/switch newtype    # Switch to newtype-profile
/switch omo        # Switch to oh-my-opencode
/switch none       # Disable all plugins

Note: After switching, you need to restart OpenCode (Ctrl+C, then opencode).

The first time you use /switch, it automatically installs itself to ~/.config/opencode/command/switch.md. This means the command remains available even after switching to other plugins (like oh-my-opencode), allowing you to switch back anytime.

Differences from oh-my-opencode

Aspect	oh-my-opencode	newtype-profile
Scenario	Code programming	Content creation
Main Agent	Sisyphus	Chief (Editor-in-Chief)
Sub Agents	oracle, librarian, explore...	researcher, writer, editor...
Categories	visual-engineering, ultrabrain...	research, writing, editing...
Tool	sisyphus_task	chief_task

License

This project is based on oh-my-opencode and follows its SUL-1.0 License.

Acknowledgments

• oh-my-opencode - Original project
• OpenCode - AI programming platform
• Google Antigravity - Model authentication