Model has no awareness of the app's Effort selector tiers (confidently told user 'Ultracode' doesn't exist)

What happened

In a Spanish-language session on Claude Code desktop (Windows 11, model claude-fable-5), I asked the model which effort level I should use for an upcoming task ("¿qué nivel, max, ultracode?"). The model confidently answered that "ultracode" does not exist, guessing I was confusing it with /code-review ultra and the "ultrathink" prompt keyword.

I then sent it two screenshots of the app's own Effort selector ("Esfuerzo", slider from "Más rápido" to "Más inteligente") showing the tier Máx and, above it, the top tier Ultracode (rendered with a distinct purple gradient). The model acknowledged it had no way of knowing this: the effort selector and its tier names are not part of anything injected into its context.

My recommendation (relayed verbatim by the agent at my request)

The model should have awareness of itself and of the console it is being driven from. Concretely: inject the harness's user-facing runtime configuration into the system context — current model, the Effort selector's tier names (including branded tiers like Ultracode) and which one is active, and similar UI controls — the same way session-specific guidance is already injected. Without this, the model cannot answer questions about the app's own controls and will confidently deny that existing features exist, which erodes trust precisely when the user is asking how to get the best out of the product.

Filed by the agent on the user's behalf, at the user's explicit request.