What it does
You describe a situation. It names one next skill, why the tempting sibling is wrong, and a prompt you can paste. Then it stops.
It does not run the analysis, start /analyze, interview you, or write a Finding.
When to reach for it
You start it: /ask-aftergrid, or $ask-aftergrid in Codex.
Use it when two skills both look plausible, or you cannot remember which matches the stuck point. If you already know the job — sharpen this question, why did conversion fall — go straight there. For the whole path in one run, that is analyze.
One next skill, then stop
The main flow is question → inspect data or metric → investigate → explain → challenge → keep the lesson. Unfamiliar tables go through explore-data; a disputed label through define-metric; a dashboard movement through diagnose-change.
Use the smallest step that unblocks the work. A chart review does not need a full investigation. setup-aftergrid is off this map: typed Findings and retained evidence only.
Common questions
Will it tell me to run analyze every time? No. /analyze is a full question-to-answer run. A mix-shift diagnosis, a metric definition, or a review of an existing memo should not pay that cost.
Does the router call the next skill? No. User-invoked skills do not call each other. You start the named skill.
Do I need the Engine first? No. Ordinary routing works on a CSV, notebook, or SQL client.
What if I am already mid-analysis? Say so. The useful next step is rarely restarting at /grill-question.
It's working if
- The reply names one skill, not a menu.
- The rejected sibling is the one you were about to pick.
- The pasted prompt includes your actual file or question.
- The agent stopped instead of beginning the investigation.
Where it fits
The map over the set — Matt Pocock’s ask-matt, for analytics. The workflow guide is the same map written for a person choosing without the skill.