ccg

Skill: ccg

Coordinate Claude, Codex, and Gemini models for optimal results.

When to Use

• Complex tasks benefiting from multiple perspectives
• Code requiring different model strengths
• Research combining reasoning and generation
• When single-model approaches are insufficient

Model Strengths

Claude

• Complex reasoning and analysis
• Nuanced understanding
• Writing and editing
• Architecture and design

Codex

• Code generation and completion
• API integration
• Boilerplate and patterns
• Technical precision

Gemini

• Research and information synthesis
• Multi-modal processing
• Long-context analysis
• Creative exploration

Routing Strategy

ccg:claude-first

Use Claude for primary work, offload to Codex/Gemini as needed.

• Best for: Architecture, complex logic, writing

ccg:codex-first

Use Codex for code generation, Claude for review.

• Best for: Large-scale code generation, migrations

ccg:gemini-first

Use Gemini for research, Claude for synthesis.

• Best for: Exploring new domains, information synthesis

ccg:round-robin

Rotate through models for different phases.

• Best for: Complex multi-phase tasks

Execution Flow

┌─────────┐     ┌─────────┐     ┌─────────┐
│ Claude  │ ->  │ Codex   │ ->  │ Gemini  │
│ Phase 1 │     │ Phase 2 │     │ Phase 3 │
└─────────┘     └─────────┘     └─────────┘
      ↓              ↓              ↓
      └──────────────┴──────────────┘
                     ↓
              ┌─────────────┐
              │ Integration │
              └─────────────┘

Output Format

## CCG: {task}

### Model Assignment
| Phase | Model | Task | Outcome |
|-------|-------|------|---------|
| 1 | Claude | {task} | {summary} |
| 2 | Codex | {task} | {summary} |
| 3 | Gemini | {task} | {summary} |

### Execution
- **Strategy:** {strategy}
- **Total duration:** {time}

### Synthesis
{how outputs were combined}

### Final Result
{summary}

Constraints

• Match model to task strength
• Avoid unnecessary model switching
• Clear interface between phases
• Document model-specific limitations
• Budget token usage across models