Weekly review (Hermes edition)

The daily notes already say what happened. A weekly review that repeats them is worthless. Its job is to say what the week meant — the pattern across days that no single day shows.

1. sb_brief, then sb_vault_activity(since_hours=168) for everything the week touched.
2. sb_read each 01-daily/ note of the week. Read them all before writing anything: the pattern is only visible once.
3. sb_agenda() for the state the week ends in, and sb_maintain(scope="staleness", apply=false) for the relationships that cooled while the week was busy.
4. Synthesize, in the working language. Three or four themes, not ten. For each: what happened, what it cost, what it means. A theme that is just a list of meetings is not a theme.
5. Cover explicitly, because these are what a week hides:
   • Decisions made and, more usefully, the ones deferred again. Name what deferring costs.
   • People — who took most of the week, and who should have but did not.
   • Patterns — what repeated. Compare with the previous weekly note if one exists (sb_search).
   • Energy — from the braindumps' energy field, what drained and what did not.
   • Weak signals — things mentioned once that would matter if true.
6. Plan next week in at most five items, each tied to a theme. Not a task list: /prioritize does that.
7. sb_create_note(type="daily", fields={date, tags:["weekly"]}, preamble=<2-3 English sentences: what week this covers, why, when>, sections={...}, slug=<"YYYY-WNN weekly">), then sb_daily_append(section="Thinking", line="- Weekly review: [[<path>]]").
8. sb_commit("weekly-review: <week>").
9. Report the review itself in the reply, not just its path — for the scheduled run, the reply is the deliverable. Keep it to one screen.