wtf is gpt5.6-soltelu-model-inverter?

This project is a skill for GPT Codex, GPT Work, and Codex CLI that automatically picks which GPT 5.6 model variant and how much reasoning effort to use for a task, based on how demanding that task is. The three model variants it chooses between are called Luna, Terra, and Sol, ranging from a lighter and cheaper option up to the most capable one. The goal stated by the author is not to save the fewest tokens possible, but to find the best balance between answer quality, reasoning cost, and the risk of having to redo work. Simple, mechanical, low risk tasks are routed to Luna at low or medium reasoning effort. Regular implementation work and bounded debugging go to Terra at medium effort. Tricky changes that touch several parts of a codebase go to Terra at high effort. Architecture decisions, conflicting evidence, security work, data handling, or anything with real production risk go to Sol at high effort. The tool only chooses among these three GPT 5.6 variants and will never quietly fall back to an older model. When working inside Codex, if the currently active model does not clearly match the route the tool selected, it creates a separate, visible copy of the current task in the same folder using the correct model setting, so the user can keep working with the right model rather than silently continuing on a mismatched one. Each subtask handed off to a helper agent is evaluated on its own, so a simple parent task can still hand a security sensitive piece of work to the strongest model, and the reverse is also true. The most powerful setting, called Ultra, is never picked automatically and only runs if a user asks for it directly. To install it, you clone the repository and run the provided install script, which can also be pointed at a custom Codex home folder or run in a preview mode that changes nothing. A test script is included to validate the installed skill package.