As a user, when I change the model name in a --replay --edit, such as when trying to downgrade the model after a refusal, I expect the edited configuration to be parsed and applied.
# Active Configuration
> NOTE: You can edit this configuration to apply it to the current conversation.
> If the configuration is invalid, the editor will re-open.
>
> Only the most relevant configuration is shown here, but you can add any
> configuration properties you want to apply.
```toml
[assistant.model.id]
provider = "anthropic"
name = "claude-opus-4-8"
[assistant.model.parameters]
reasoning = "auto"
```
Analysis: this is a delayed-application error that you could call an off-by-one error. The state is stored but the previous state is applied. A second --replay --edit with no edits will apply the model change.
As a user, when I change the model name in a --replay --edit, such as when trying to downgrade the model after a refusal, I expect the edited configuration to be parsed and applied.
Analysis: this is a delayed-application error that you could call an off-by-one error. The state is stored but the previous state is applied. A second
--replay --editwith no edits will apply the model change.