/grill: OMP asks the right questions before coding

An agent working from a wrong assumption codes fast… in the wrong direction. Before a single line gets written, the real questions are better settled. Since Soreno 0.7.0, Chat (OMP) has a command for that: /grill followed by the subject.

One question at a time, with its recommendation

OMP starts by exploring the code: whatever the project can tell it, it doesn’t ask you. Then it asks its questions one by one, the most structuring first. Each comes with two to four concrete options, the one it recommends and, for each, the reason and what it costs. It modifies no file during the interview.

The current question on top, the decisions taken below.

The Framing column

The interview opens in a column to the right of the conversation. On top, the current question: follow the recommendation, pick another option or answer in your own words. Below, the decisions taken, each marked “Recommended” or “Your answer”. They are read back from the history: resume the conversation tomorrow and the framing is still there.

Changed your mind about a decision? Revisit puts it back on the table. And the column can be resized, or hidden to leave the conversation the whole width.

A summary at the end

When no material decision is left open, OMP says so and stops asking. Summarize then asks it for the wrap-up in Markdown: the goal, each decision and its rationale, the open questions and the next steps. Copy summary grabs it, for an issue or to start the implementation right away.

And for questions you need to see

Some questions are better settled by looking than by reading. Mock up this question asks OMP for an HTML mockup with one variant per option, then asks you the question again: that’s what the /mockup post is about.

The interview is inspired by the grill-me skill, with nothing to install: OMP in the project’s environment is all it takes.