SendMessage resume of a subagent drops the model override passed at spawn

## Environment

• Claude Code 2.1.205 (Windows 11, desktop app entrypoint)
• Session model: claude-fable-5

Repro

1. Spawn a subagent via the Agent tool with an explicit model override, e.g. Agent(subagent_type: "general-purpose", model: "opus", prompt: ...).
2. The subagent's run ends (in our case it ended prematurely after a skill auto-invocation, see #76385, but any completed run reproduces this).
3. Continue the agent via SendMessage to its agentId. The tool reports: Agent "<id>" had no active task; resumed from transcript in the background with your message.

Observed (subagent JSONL transcript)

• First-run assistant messages: "model": "claude-opus-4-8" (override respected).
• All post-resume assistant messages: "model": "claude-fable-5" (the session model). The spawn-time model: "opus" override is silently dropped.

Expected

Either the resume should preserve the model override passed at spawn, or SendMessage should accept a model parameter for the continuation. At minimum the behavior should be documented, since it silently changes cost and capability mid-task.

Impact

Orchestration patterns that deliberately route execution work to a cheaper/different model (session model plans and reviews, subagents execute) lose the override on any follow-up message; the cost difference is invisible unless one inspects the transcript.

Workaround

Do not resume a failed execution agent via SendMessage; spawn a fresh agent with the explicit model instead.