Model selection scoped to a single task got applied to the whole session, causing unexpected token/credit consumption

Description

Requested Fable 5 for a specific task within a session, not as the session-wide default. Instead the session ran on Fable 5 throughout, consuming token usage well beyond expectation and exhausting credits mid-session — hit "You're out of usage credits. Run /usage-credits to keep using Fable 5 or /model to switch models." during a /compact call.

Expected behavior

A per-task model request should not silently become the session-wide model selection.

Actual behavior

The entire session ran on the higher-cost model, burning through token/credit budget unexpectedly.

Impact

Exhausted usage credits mid-session. Requesting review of the usage this caused and consideration of a credit adjustment.