Generate a plan using multiple models

[tomzx]

When generating a plan, we should support having multiple different models do it (e.g., OpenAI models, Anthropic models, Google models, etc.).
With this approach we may benefit from the variety present in the different trained model and where they put their attention.

Similarly, it may make sense to generate plans using the same model but with different prompts, for instance based on different developer roles.

[github-actions]

Implementation Plan Generated

I've created a detailed implementation plan for multi-model plan generation.

📋 Plan PR: #5

Overview

The plan proposes a comprehensive multi-model orchestration system with:

• Support for multiple providers (OpenAI, Anthropic, Google)
• Three strategy types: diverse (different providers), role-based (different perspectives), consensus (multiple models)
• Parallel execution using GitHub Actions matrix strategy
• Intelligent LLM-powered aggregation of outputs

Complexity

This is a high complexity implementation requiring:

• Configuration system for models and strategies
• Provider adapter scripts for OpenAI and Google
• Workflow restructuring with parallel execution
• Robust plan aggregation logic
• Gradual rollout with feature flags

Key Design Decisions

1. Parallel vs Sequential: Run models in parallel for speed
2. Aggregation: Use Claude Opus to synthesize multiple outputs
3. Backward Compatibility: Keep single-model workflow as fallback
4. Cost Management: Configurable strategies to control API usage

Please review the plan and provide feedback on:

• Which strategy should be default?
• Budget/cost constraints?
• Provider priority for implementation?

Once approved, the implementation can begin following the 6-phase approach outlined in the plan.