problem history
[Anthropic Messages API, Claude 4.5+] input + max_tokens over the context window is accepted, then output stops with stop_reason 'model_context_window_exceeded' — agents that only check 'max_tokens' …
Revision reasons are not recorded in this slice.