[EP3] oh-my-opencode vs MoAI-ADK - Enhancement Layer In-Depth Analysis

Analysis Overview

1. Basic Architecture Comparison

2. ToS and Stability Comparison

2.1 January 2026 Anthropic Technical Restriction

🚨 CRITICAL UPDATE

Error message: "This credential is only authorized for use with Claude Code."

Impact:
├── Cannot use Claude Code OAuth tokens in third-party tools (including OpenCode)
├── Multiple account suspension cases reported when using OpenCode with OAuth subscription
└── Recommended workaround: Use Anthropic API key directly (incurs additional costs)

"START WITH YOUR ChatGPT, Claude, Gemini SUBSCRIPTIONS. WE ALL COVER THEM."

3. Agent Ecosystem Comparison

3.1 oh-my-opencode Team Composition (6 Specialized Agents)

Sisyphus (Claude Opus 4.5, 32K tokens) - Orchestrator
├── Oracle (GPT-5.2): Strategic reasoning and debugging
├── Librarian (Claude Sonnet 4.5 / Gemini 3 Flash): Documentation and analysis
├── Frontend UI/UX (Gemini 3 Pro): UI/UX development
├── Document-Writer: Documentation specialist
├── Multimodal-Looker: Image/screenshot analysis
└── Explore: Fast codebase pattern matching

"Recently rewrote an old PostgreSQL-compatible SQL layer with oh-my-opencode."

"Processed 8,000 ESLint warnings in one day."

"Converted a 45,000-line Tauri app to SaaS overnight."

3.2 MoAI-ADK Agent Ecosystem (20 agents)

Alfred Orchestrator
├── Manager Agents (8)
│   ├── manager-git: Commit, branch, PR management
│   ├── manager-spec: EARS format requirements, acceptance criteria
│   ├── manager-tdd: RED-GREEN-REFACTOR cycle
│   ├── manager-docs: README, API docs, technical documentation
│   ├── manager-quality: TRUST 5 validation, code review
│   ├── manager-project: Project setup, configuration
│   ├── manager-strategy: Architecture decisions, technology evaluation
│   └── manager-claude-code: Claude Code configuration, MCP integration
├── Expert Agents (8)
│   ├── expert-backend: API design, authentication, database
│   ├── expert-frontend: React, Vue, Next.js, components
│   ├── expert-security: OWASP, vulnerability assessment, secure coding
│   ├── expert-devops: CI/CD, Docker, Kubernetes
│   ├── expert-performance: Profiling, optimization, benchmarking
│   ├── expert-debug: Error diagnosis, troubleshooting, root cause analysis
│   ├── expert-testing: E2E, integration testing, QA automation
│   └── expert-refactoring: AST-based transformation, large-scale changes
└── Builder Agents (4)
    ├── builder-agent: Custom agent creation
    ├── builder-command: Slash command development
    ├── builder-skill: Knowledge skill authoring
    └── builder-plugin: Marketplace plugin development

4. Core Feature Comparison

5. Execution Capability Comparison

5.1 oh-my-opencode: "Relentless Execution"

• Forces agents not to give up midway
• Unlimited execution until goal completion
• Completion detected by <promise>DONE</promise> marker

5.2 MoAI-ADK: "Checkpoint-Based Execution"

1. AskUserQuestion Checkpoints: User confirmation at major decision points
2. Type A Command Constraints: Workflow commands allow Task only
3. TRUST 5 Quality Gates: Progress blocked when quality criteria not met
4. Git Checkpoints: Rollback possible with automatic backups
5. Subagent Isolation: Limited permission scope for each agent
6. Configuration-Based Control: Customize behavior via .moai/config/
7. Resumable Patterns: Resume interrupted work via agentId
8. Output Style Enforcement: Consistent response styles like R2-D2, Yoda

oh-my-opencode Execution Model

MoAI-ADK Execution Model

5.3 User Control

6. Context Efficiency Comparison

6.1 oh-my-opencode

• Preemptive Compaction: Automatic summarization at 85% context usage
• Extract only necessary information through multimodal analysis
• Directory-Specific Context Injection (automatic AGENTS.md collection)

6.2 MoAI-ADK

• 200K Token Budget Management: 7.5% system + 75% conversation + 10% reference + 7.5% reserve
• /clear recommendation system (token prediction + automatic guidance)
• Progressive Disclosure (5-Tier loading)
• Preemptive Compaction: Automatic summarization at 85% context usage
• Configuration modularization (sections/*.yaml)
• Reduction Rate: 60-70%

7. Workflow Philosophy Comparison

7.1 oh-my-opencode Automatic Task Decomposition and Team Allocation

7.2 MoAI-ADK Workflow: /moai:alfred One-Click Automation

Intelligent Routing

3-Phase Workflow

Quality Gate Checkpoints

Multi-LLM Mode

Interrupt/Resume Support

# Resume interrupted workflow
/moai:alfred SPEC-AUTH-001 continue please!!

# State storage location: .moai/cache/alfred-{spec-id}.json

Usage Examples

# Basic usage
/moai:alfred "User authentication with JWT tokens"

# Automatic branch creation
/moai:alfred "Shopping cart feature" --branch

# Automatic PR creation
/moai:alfred "Payment integration" --branch --pr

7.3 Quality Assurance

8. Selection Guide

When oh-my-opencode is Suitable

1. When preferring autonomous execution: When wanting work to proceed without intervention after starting
2. When wanting to utilize various models: When wanting to switch and use GPT, Claude, Gemini according to situation
3. When rapid prototyping is needed: Early development stage where speed is more important than quality

When MoAI-ADK is Suitable

1. When ToS compliance is important: When wanting stable use without account suspension risk
2. When quality-focused development is needed: Enterprise environment where TDD, code review, quality gates are essential
3. When wanting controlled automation: When balance between AI's strong execution capability and user control is needed
4. When conducting long-term projects: When stable tool support and consistent quality are important

Next Episode Preview

1. /moai:alfred: 759-line one-click development automation command
2. Ralph Engine: LSP + AST-grep + Loop Controller integrated quality engine
3. Anti-Hallucination Strategy: 90+ domain skills system

References