Check a live campaign

Read-only. This command cannot change anything on Meta, by design: the thing that watches the spend should not also be able to change it.

Run it

PYTHONPATH="${CLAUDE_PLUGIN_ROOT}" python3 -m adscooking pulse

Exit code is the verdict:

Exit 3 is deliberately separate. A missing .env is not a reason to tell someone their ads are in trouble.

What to tell the user

Match the reply to the exit code. Do not pad a quiet day into a paragraph.

Exit 0. One line. "$41 spent, 2 leads, $20.50 each. Nothing to flag." Stop there.

Exit 1. The headline numbers, then the specific warning and what it means. Still short.

Exit 2. Lead with the problem, not the numbers. Say what it is, what it costs if ignored, and the one thing to do about it. If they asked for a background or recurring check, this is the case worth interrupting them for.

Reading the findings

Cost per lead above the band. The band in their config is a starting point, not a target. Before suggesting anything, check how many leads it is based on: under about 3 it is noise, and the tool says so rather than giving a verdict.

Spend with no leads. These thresholds are house rules, not research. They mean "you have now spent enough to know something is wrong", not "Meta is broken".

An edit was detected. Someone changed the live ad set since the last check. Editing a live ad set restarts the learning phase, which at a small daily budget can cost several days of delivery. If it happened inside the no-edit window, say so plainly.

Token expiring. Real deadline, and the tool gives you the date. Use it: "expires 14 March" is something they can act on in a way that "expires in 12 days" is not. When it dies, the ads keep running but every command here stops working, silently.

No spend today on an active ad set. Before mid-morning this is usually nothing. Later in the day it means delivery has stopped.

Nothing about edits on the first run. Edit detection works by comparing against the previous run, so the first one has nothing to compare to. That is expected, not a fault. Say so if they ask why it did not check.

What not to do

• Never change anything in response to what this reports. Not the budget, not the copy, not the status. Report, then let them decide. A run of this command that ends in an edit is a bug in your behaviour, not a helpful extra.
• Do not invent a threshold. If a number is outside what the config judges, log it and say it is not something you have a benchmark for. ${CLAUDE_PLUGIN_ROOT}/context/ad-management-principles.md lists what was verified and what was not, on purpose.
• Do not report a cost per lead verdict off one or two leads. It is the fastest way to talk someone into killing an ad that was fine.

Running it daily

If the user has the /loop skill, /loop 24h /ads-cooking:pulse runs it daily. Otherwise a cron entry or a launchd agent does the same thing; the command is a plain shell command with a meaningful exit code, so anything that can run one works.

It keeps its own history in the config folder (pulse-history.jsonl), so after a few weeks their own numbers are a better benchmark than any published median.