This is a test of T3's MCP argument validation. Do exactly these two tool calls, nothing else, then reply with each call's raw result verbatim (error text or JSON), labelled CALL 1 and CALL 2.
CALL 1: the T3 MCP tool t3_thread_launch with exactly these arguments: {"title": "Repro: prompt instead of message", "scratch": true, "prompt": "say hi"}
CALL 2: the T3 MCP tool t3_thread_launch with exactly these arguments: {"title": "Repro: message", "scratch": true, "message": "say hi"}
Do not correct, rename or drop any argument key in CALL 1, even if you expect it to fail.