| name | weekly-operating-review |
| description | Friday weekly OS health pass. Runs /audit, then /level-up, then refreshes kb/wiki/_hot.md from current state. Single command instead of three. Logs the run. Does not build artifacts. |
| disable-model-invocation | true |
/weekly-operating-review — Friday Pulse
The Friday weekly cadence in one skill. Adapted from AIS-OS weekly improvement loop.
When to use
- Friday afternoon (or whenever your week ends)
- Weekly cadence, no exceptions for 90+ days
When NOT to use
- Mid-week ("I'll just do a quick review") — defeats the trend
- Daily — that's
/daily-plan + /end-of-day-review
Inputs
- Last 7 days of
decisions/log.md
- Last 7 days of
~/.claude/usage.jsonl (if exists)
- Most recent
docs/audits/*.md
- Current
context/Priorities.md
Outputs
- A new audit report (via
/audit)
- A
/level-up recommendation logged in decisions/log.md
- A refreshed
kb/wiki/_hot.md
- One end-of-week entry in
decisions/log.md
Steps
1. Pre-flight
git status --short — flag dirty state
- Read most recent audit; note total score
- Read most recent
/level-up entry; note whether the recommended artifact was built
2. Run /audit
Invoke the audit skill. Capture: total score, top 3 gaps, recommended next action.
3. Run /level-up
Use the audit's top gap as the constraint. Walk the 5 questions. Land on one artifact.
4. Refresh kb/wiki/_hot.md
Re-derive each section:
- This week's focus ←
context/Priorities.md "this week" + audit's top gap
- Active decisions in play ←
decisions/log.md last 7 days, status Active or Planned
- Open candidates from /level-up ←
decisions/log.md Type: <skill|script|...> entries with Status: Planned and not yet built
- Hot wiki articles ← inspect
kb/wiki/log.md for articles touched in last 7 days; pick top 5 by relevance to focus
- Active connections at risk ←
connections.md rows with Status: auth pending|paused
- Leverage gap of the week ← audit's top gap (one sentence)
Write the file. Update last-updated: to today.
5. Log the weekly review
## YYYY-MM-DD — Weekly operating review
**Audit total:** X/100 (Δ from last: ±Y)
**Top gap:** <one line from audit>
**Level-up recommendation:** <artifact + why>
**Hot cache refreshed:** yes
**Status:** Reflective; no artifacts built this run.
**Follow-up:** Build the recommended artifact this coming week. Re-audit next Friday.
6. Output to user
Three-line summary:
- Score (with delta)
- Top gap
- Recommended artifact
That's it. The full reports are in their files.
Hard rules
- MUST NOT build anything. This is reflection only.
- MUST update
_hot.md even if score didn't change.
- MUST log even if no recommendation surfaced (a "stable, no recommendation" week is signal).
- MUST flag if last week's recommendation wasn't acted on.
Failure modes
| Symptom | Cause | Fix |
|---|
| Last week's artifact wasn't built | Operator was busy / didn't agree / forgot | Don't re-recommend the same thing automatically. Ask: "Last week's recommendation was X. Still relevant?" |
| Audit and last audit are identical | Either nothing changed, or /audit is broken | Inspect decisions/log.md for movement. If real movement happened but score unchanged → rubric needs adjustment. |
_hot.md ends up empty | Sources are empty (no decisions, no priorities, no audits) | This is real signal — the OS isn't being used. Surface this honestly. |
Reference files
references/four-cs-framework.md — what /audit measures
references/3ms-framework.md — what /level-up thinks about
docs/CADENCE.md — when this skill fires
docs/WIKI_LAYER.md — what _hot.md is for
Improvement loop
After each weekly review:
- Did the operator act on the recommendation? If never →
/level-up recommendations are wrong scope or wrong priority
- Is the trend line going up? If flat for 4 weeks → either the OS plateau is real, or the rubric is mis-calibrated
- Is
_hot.md actually being read? If not → either reduce its size further, or stop maintaining it