Update the no-tools chat planner prompt so questions about the owner's own life/history dispatch the read talent by default, without requiring an explicit "check the journal" cue. Non-journal turns still answer directly: greetings, acknowledgements, in-context follow-ups, opinions, general knowledge, and questions about sol's own role. This addresses req_zzrbuflb: the old "dispatching is the exception" frame caused the small local default model to answer plausibly from general knowledge instead of using the only journal path. Offline eval showed ~30% read dispatch on journal-shaped questions for that model, while capable cloud models already reached ~100%; the prompt is the intended lever for that adherence gap. Regenerate the sol preview baseline because the prompt body is embedded verbatim in full_prompt; this is expected, not nondeterminism. Preserved invariants: messages[-1] anchoring, cap=1 wording, Stop-And-Report, read-only journal-local read, and byte-identical frontmatter.