decide

Take the decisions a run has already named and close them — each one arriving as a select dialog with concrete options, rather than as a paragraph somebody has to answer by typing. One decision at a time, in dependency order, with a recommendation offered first.

The dialog is the whole contribution. The interviewing itself stays with whichever engine the repo configures; this skill supplies that engine with the decisions, forces the form every question arrives in, and carries the answers back. A wrapper that started asking questions of its own would be a second interview competing with the one the repo already chose.

Where this sits next to grilling. grilling generates questions — it probes freely and adversarially, and widens until nothing is left unexamined. This skill closes questions that already exist: it takes a list that is fixed before the run starts, and the run is over when that list is answered. Same territory, opposite direction. grilling is not this repo's skill, so the boundary is drawn here alone: the reading of it above is decide's, and what the line binds is this skill's own limit — grilling is under no obligation to know the line exists.

Manually invoked, never inside a drain. The entire product is a question a human answers, so it has no meaning unattended — the work-loop queues remove the dialog tool outright, and this skill lives on the other side of that line: somebody types /decide.

It is not itself an interview engine. grillWith names the skill that conducts an interview; this one is a caller of that skill, so it never belongs in that key. A repo that names it there has configured a loop.

Workflow

1. Collect the decisions

The input is the session, not a tracker. What this skill closes is a list that already exists — a refine-issue run that ended still open, a plan's open questions, a gap or findings report, a design note that named its trade-offs. Take them as the run named them, in the caller's own words, and add none.