Sol-Advisor: A Codex Workflow with Sol for Orchestration, Terra for Complex Execution, and Luna for Routine Execution

One-sentence takeaway

The core concept of Sol-Advisor is using Sol High for orchestration, Luna Max for routine task execution, Terra Max for complex execution, and a fresh Sol instance for review—concentrating high-capability models strictly on planning and acceptance.

Suitable use cases

• Suitable tasks: Engineering tasks in Codex that can be broken down and require a plan-execute-review closed loop.

• Unsuitable tasks: Small one-off tasks, tasks without verifiable acceptance criteria, and tasks involving destructive write operations without human confirmation.

• Applicable model versions: GPT-5.6 Sol High, GPT-5.6 Luna Max, GPT-5.6 Terra Max.

• Applicable clients, agents, or APIs: The original post refers to a Codex plugin/workflow; installation commands do not appear in the publicly visible page text.

• Recommended reasoning tiers and parameters: Sol High for orchestration/review, Luna Max for routine tasks, and Terra Max for complex tasks. This is the role division proposed by the author, not a benchmarked optimum.

Copy-ready content

Sol-Advisor Role Breakdown:
1. GPT-5.6 Sol High acts as the orchestrator: breaks down tasks, dispatches subtasks, and maintains overall objectives.
2. GPT-5.6 Luna Max acts as the routine task executor: handles well-defined, repetitive, and low-risk work.
3. GPT-5.6 Terra Max acts as the complex task executor: handles work requiring deeper reasoning.
4. A fresh GPT-5.6 Sol instance acts as the reviewer: inspects results, tests, and validates whether original objectives are met.

How to use

1. Have Sol High decompose the objective into independently verifiable subtasks.

2. Route clear, repetitive subtasks to Luna Max; route complex implementations or critical judgments to Terra Max.

3. Collect execution artifacts, test results, and unresolved issues.

4. Have a fresh Sol instance review diffs, tests, and acceptance criteria; if checks fail, route back to the executor for fixes.

5. Require human confirmation for external side effects, deletions, and deployment actions.

Visibility and boundaries in the original post

• The original post explicitly discloses the role division, but the installation instructions for the "free open-source Codex plugin" are not detailed in the currently visible text; do not invent installation commands or repository URLs.

• The original post does not provide control groups, failure rates, context usage, quota consumption, or cost figures under identical tasks; therefore, this is a workflow rather than a reproducible benchmark.

• The statement that "Luna Max on high reasoning roughly equals Sol on medium reasoning at roughly one-sixth the cost" comes from another experiential post cited by the author; this is an individual claim and should not be taken as a performance guarantee for Terra.