Two Models, One Skill: Claude Plans, Sol Codes

Where this fits

You're refactoring a core module in your agent codebase. You ask Claude Code for a plan, but you want a different model to write the actual diff—because the model that writes code shouldn't grade its own work. Sol Skill embeds this loop into Claude Code: Claude drafts a detailed brief, delegates implementation to GPT-5.6 Sol via the OpenAI Codex CLI, then reviews the resulting diff for correctness and style.

Why agents benefit

• Separation of concerns: Planning, implementation, and review are handled by distinct models, reducing self-review bias and catching blind spots.
• Cross-model verification: The reviewing model (Claude) isn't attached to the implementation choices, so it can critically evaluate edge cases and API usage with fresh eyes.
• Leverages Codex CLI: Directly taps into the OpenAI Codex command-line interface for execution, making the delegation seamless and scriptable.
• Works inside Claude Code: No need to switch tools—everything is managed via a simple /sol skill invocation within your existing agent workflow.
• Reusable pattern: Even beyond this specific skill, the idea of orchestrating multiple frontier models in a pipeline is a template you can adapt to other tasks.

Practical scenarios

Large-scale refactoring

You have a messy legacy module that needs restructuring. Claude analyses the codebase and produces a step-by-step refactor plan, Sol executes the changes, Claude reviews the diff for regressions and consistency.

Generating boilerplate microservices

You need to scaffold a new service with CRUD endpoints, database models, and tests. Claude writes the spec, Sol generates the boilerplate, and Claude verifies that the generated code matches the spec and follows your team's conventions.

Independent code review of GPT-generated code

You've written code using GPT-5.6 elsewhere but want a second opinion. Use Sol Skill to have Claude review the diff and provide a structured critique, separate from the original implementation context.

Add it to your agent workflow

Install the skill with:

npx skills add ozankasikci/sol-skill

Then invoke it from Claude Code with a prompt like:

/sol Refactor the auth middleware to use the new token validation API. Keep the public interface unchanged.

Claude will produce a brief, hand off to Sol via Codex CLI, and then review the result.

Compare before adopting

Before fully committing, consider these points:

• Review quality: Does the Claude-review step actually catch meaningful bugs, or is it just a rubber stamp? Test on real PRs.
• Maintenance freshness: Check the repo’s last commit and open issues. A skill that’s not updated for new Claude or Codex versions can break silently.
• Alternatives: You could write your own multi-model loop or use a single model with forced self-critique. Weigh the overhead of two API calls against quality gains.
• Workflow fit: Does the skill respect your existing hooks, ignore rules, and MCP config? The README and source should show how it integrates.

Why it is worth tracking

Sol Skill has only 26 stars but is structurally interesting: it implements a separation-of-duties pattern that many teams are starting to explore. The author is actively sharing it via Agent Skills and the core mechanism—Claude plans, Sol codes, Claude reviews—is a design worth watching. If it gains traction and community contributions, it could become a reference for orchestrating multiple models in agent workflows. Track it on its OpenAgentSkill page to see real-world feedback, issues, and version updates. It’s worth an evaluation whenever you’re building a coding agent that could benefit from another model’s perspective.