weekly-account-os: the Monday button

Run the week Cowork-ready

What it does

This is the Monday button. One command runs the full weekly routine on a client account: the client-facing performance readout, the full Meta account audit, and the creative scale/kill board. Then it merges all three action lists into one list, ordered by what actually costs you money, with every item citing where it came from.

It is an orchestrator, not another opinion. It sequences skills you could run individually, reads their gates, and refuses to paper over a component that could not run.

Nothing it proposes touches your account. Every account change leaves here as a nomination and goes into meta-scaling-copilot’s approval lane, where you approve or reject each individual change.

When to reach for it

Monday morning, on every account you manage. That is the design case.
You are about to get on a client call and need the week in one place.
You are managing enough accounts that running four skills each, by hand, is the thing that quietly stops happening.
You want one action list, not three that partly contradict each other.

Where it runs

Cowork ✓ Desktop ✓

Works in both. It needs your Meta Ads connector authorized to read account data. Without it, the run degrades honestly and tells you exactly which sections it could not fill.

How to run it

“Run the weekly account OS for examplebrand”

Or by name:

“/um-toolkit:weekly-account-os examplebrand”

What it needs from you

Which account. It confirms the ad account name as Meta itself reports it, next to the brand you said, before any verdict. That check exists because a near name-match is otherwise indistinguishable from the real thing.
A connected Meta Ads account (read-only is enough).
Your money-path numbers where you have them: store or CRM revenue is what the verdicts are graded against, never Ads Manager ROAS.

What you get back

One consolidated action list, leak-ranked, each item citing its source skill and a dated decision rule.
The client-facing readout, wins and losses both stated plainly, ready to send.
The creative board: scale / iterate / kill / hold, with the spend floor that earned each verdict.
Account nominations queued for your per-change approval.
A cost ledger for the run, because this fires 50+ times a year per client.

Deliverability rides along as a monthly row. drop-triage fires automatically if the week shows an unexplained sudden drop.

A worked example

You: “Run the Monday routine on examplebrand.”

It returns: Account confirmed as Example Brand LLC (act_1234…). Spend up 12%, banked revenue flat, so aMER slipped. #1 leak: three ad sets sitting under the learning floor are eating 22% of spend with no verdict possible yet, cited to the dated learning rule. Creative board: 2 scale, 1 kill past the $500 floor, 4 still under floor and explicitly not judged. Queued: 3 budget nominations awaiting your yes.

Tips & gotchas

It never writes to your account. If you want a change made, you approve the specific diff. There is no bulk approve.
A component that could not run shows as NO-DATA, and NO-DATA blocks like a failure does. A week where the dataset was unreachable will not come back looking like a clean week. That is deliberate.
Under-testing gets flagged too. A suspiciously high hit rate means you are not testing enough, and the board says so.
Run it on the same day each week. The anomaly detection compares against trailing 8-week bands, so a consistent cadence makes the bands meaningful.

It sequences weekly-readout, meta-account-doctor and creative-testing-os, routes changes through meta-scaling-copilot, and calls drop-triage and deliverability-doctor when their triggers fire.