Why switching models can lose access to skills

The reported failure

Users run Claude through T3 Code and see their existing skills work. 40:30
They switch to Codex and find that those skills are missing, then look for a T3 Code skills setting. 40:30
Theo asks whether they configured the skills in Codex. Their expectation was that changing the model would preserve everything automatically. 41:06

Theo says T3 Code's apparent model picker is a harness picker. Moving between providers can change the harness, with its own tools, skills, plugins, and configuration. 41:29 41:52
Adopting T3 Code replaces the human interface around a bundled model-and-harness product. It does not automatically transfer one harness's configuration into all the others. 42:26
His analogy is a food court: one interface offers access to several kitchens, but selecting a different kitchen changes what it can supply. 41:29

Theo sees the confusion as evidence that the abstraction is working: users have stopped thinking about the underlying harness. That also makes consistent access to capabilities T3 Code's problem to solve. 41:06 41:52 42:54
His desired experience is to switch harnesses as readily as changing a model identifier, without relearning the interface whenever the best available model changes. He presents this as a direction for the product, not a claim that all configuration already transfers. 42:54 43:19