To questionnaire

Adapted from mattpocock/skills to-questionnaire (MIT). Companion to planning/grill: grilling mines the user's own head, this mines someone else's.

Run it in the same conversation as the grilling session that stalled — the open questions are already in context. Started fresh, it knows nothing about the subject and the user has to re-supply it.

Interview — two questions, then stop

1. Who is it going to? Role, expertise, relationship to the user. Fixes tone and how much context the document carries — an outside client needs orienting, a teammate doesn't.
2. What do you need back? The concrete decisions or facts the user can't resolve alone. This is the checklist the finished document is measured against: every item named gets a question aimed at it.

Do not interview the user about the subject. Not knowing the subject is the whole reason for writing to someone else.

The document — to-questionnaire-<slug>.md in the working directory

Framed as a discovery questionnaire (the user lacks the context, the recipient holds it):

• Purpose line naming the decision riding on it, then a short context section for a recipient who was never in the user's head.
• Questions ordered most-important-first, grouped under themed headings — async means one pass may be all you get.
• One idea per question, never compound. An answer stub beneath each. A why this matters line only where a question could be misread.
• Explicit permission to answer "I don't know" — a flagged uncertainty is useful, a confident guess that reads like a fact is not.
• Closing catch-all: "anything we didn't ask that we should know?"

Not branching (flat grouped list, not a tree). Not multi-recipient — one run, one document, one person. If three people hold three parts, run it three times.

After

The answers come back as raw material — feed them into the next planning/grill round, or into planning/write-prd if the work is heading for a build.

Anti-patterns

• ❌ Asking the user a question about the subject itself — the skill is off the rails.
• ❌ Compound questions, or questions the recipient can't answer without the user's context.
• ❌ Copying the user's open questions down verbatim instead of re-aiming them at what the recipient knows.
• ❌ One document trying to serve several recipients.
• ❌ Starting it in a fresh session with no grilling context to draw on.