/weekly-review

name: weekly-review
description: Generate the weekly review report for the current ISO week. Writes to .hermit/compiled/review-weekly-YYYY-Www.md (dev-facing detail) and sends a plain-language channel summary (Delivered / Decisions / Waiting on you / Spend). Runs every Sunday at 23:00 via routine.

Record notes only inside an open record's turn, using `bun ${CLAUDE_PLUGIN_ROOT}/scripts/task.ts note .hermit <id>` with the note on stdin. Otherwise skip record notes. Never edit a task file directly.

Weekly Review

Generates the weekly review for the current ISO week.

Commands

• `duties-summary`: `bun ${CLAUDE_PLUGIN_ROOT}/scripts/duties.ts summary .hermit`
• `task-report`: `bun ${CLAUDE_PLUGIN_ROOT}/scripts/task-report.ts .hermit --recent --limit 3`
• `artifact-render`: `bun ${CLAUDE_PLUGIN_ROOT}/scripts/artifact.ts render <type> .hermit`

Steps

1. Run:

   bun ${CLAUDE_PLUGIN_ROOT}/scripts/weekly-review.ts .hermit

2. Report the review's `path`. When `knowledge_findings` is non-zero, summarize the review's **Knowledge Health** section to the operator; when `usage_docs` is, relay its **Usage** section: the script has already auto-archived the docs with no tracked use in the window (move-only, into `compiled/.archive/` — restoring one is moving the file back, and a restored doc is never archived again). Don't archive anything further yourself, and never on judgment alone: semantic changes (merging or rewriting topic pages, reclassifying a compiled conclusion as raw, tagging `foundational`) always wait for explicit operator approval. Tracked use for this section is compiled/ Reads including subagent reads; startup injection isn't tracked.

3. **Dispatch the week's topic semantic check** to `hermitd:skill-eval-runner` once, passing no `model` parameter so it inherits the session model. Point it only at `${CLAUDE_PLUGIN_ROOT}/skills/weekly-review/reference.md`, which reads topic pages for contradictions, stale claims and broken wikilinks, capped at three findings. Copy the Commands entries verbatim into the dispatch prompt. Run `bun ${CLAUDE_PLUGIN_ROOT}/scripts/memory-dir.ts` and name the returned absolute `dir` for resolving memory links, plus the current ISO week from step 1. The check only reports; semantic rewrites from its findings still require operator approval.

The runner returns:

<!-- weekly-review-eval-schema:start -->

{
  "topic_findings": [ "<one-line finding>" ]
}

<!-- weekly-review-eval-schema:end -->

**Failure policy:** on null or malformed JSON carry `topic_findings: []` and continue. Show `Topic pages:` in the channel summary only when findings are non-empty.

4. Prune channel-log rows by age, independently of the nightly transcript maintenance:

   bun ${CLAUDE_PLUGIN_ROOT}/scripts/channel-log.ts .hermit prune <knowledge.channel_log_retention_days from config.json, default 90>

A genuine DB failure is fail-open: report it and continue with the review.

5. Read the frontmatter needed for the channel summary from the freshly-written review file:

• Read `.hermit/compiled/review-weekly-<current-week>.md` frontmatter (just written in step 1) — do not read the body; every value the channel message needs lives in frontmatter (`delivered_count`, `delivered`, `proposals_accepted`, `proposals_resolved`, `open_loops_count`, `total_cost_usd`, `usage_untouched_count`, `usage_auto_archived`, `skills_new`, `skills_archived`, `skills_often_patched`).
• Also read the prior week's `compiled/review-weekly-*.md` frontmatter (sort by `week` descending, take the second file) for the Spend delta.
• If no prior week file exists: omit the "vs prior week" comparison and show this week's spend only.
• If the current-week file is missing (script failed): skip step 6 entirely and fall back to a plain note ("Weekly review didn't generate this week — nothing to send.").

6. **Channel voice rule:** the message below is for the person who owns this hermit, not a developer. Never emit task ids (`T-...`), raw token counts, cron strings, or file paths. Speak in plain outcomes and counts.

Channel-send the combined weekly summary:

• Refresh the dashboard per `${CLAUDE_PLUGIN_ROOT}/docs/artifacts.md`; if it returns a URL, note it for the message below.
• Publish the weekly-review artifact when `config.artifacts.weekly_review` is on: render it with `bun ${CLAUDE_PLUGIN_ROOT}/scripts/artifact.ts render weekly .hermit` (the page id is `weekly`, not the config key), then follow the hash-gate/publish/state-write steps in `${CLAUDE_PLUGIN_ROOT}/docs/artifacts.md` under state key `weekly_review`; if it returns a URL, note it for the message below too.
• Compose the message in these sections. Show a line only when it has something to report — except Dream and Spend, which always show (spend visibility matters even at $0):

     Delivered: <delivered_count> thing(s) — <up to 5 of the delivered names, comma-joined and plain; add "and N more" for the rest> [omit this whole line when delivered_count is 0]
     Decisions: <proposals_accepted> approved, <proposals_resolved> resolved this week [omit this whole line when both are 0]
     Waiting on you: <open_loops_count> thing(s) need a yes/no [omit this whole line when 0]
     Tidied: put <usage_auto_archived count> unused doc(s) away — <up to 5 of the usage_auto_archived names, comma-joined and plain; add "and N more" for the rest> — say "restore <name>" to bring one back [omit this whole line when the list is empty]
     Unused: <usage_untouched_count> doc(s) look dormant — say "archive them" if you want them tidied away [omit this whole line when 0]
     Skills: <skills_new> new, <skills_archived> put away, <skills_often_patched> often adjusted [omit when all are 0]
     Dream: <step 1 dream.runs> runs, <dream.flagged> flagged, <dream.changes> file changes
     Spend: $<total_cost_usd> this week (vs $<prior week's total_cost_usd>, if a prior file exists) — an estimate, not a