Tune system prompt to avoid asking questions at the end of the output. Update debug thinking to include the LLM model used for that chat.
Signed-off-by: Daniel Henry <iamdanhenry@gmail.com>
