Grilling
Interview the user relentlessly until reaching shared understanding. Nothing gets silently assumed.
The plan is a design tree: every decision branches into the decisions that hang off it. Visit every branch, surfacing the decisions the user has not yet noticed they are making.
Hand off instead when
| The answers are... | Skill |
|---|---|
| Not formed yet — the plan itself has to be invented | brainstorming |
| In the user's head — a plan exists but is undocumented | grilling |
| On disk — a PR, code, or written design doc you can read | scrutinize |
Say so and stop, rather than interviewing someone about a document you could have read. To leave a glossary and ADRs behind, use grill-with-docs.
Rounds
The frontier is every decision whose prerequisites are already settled: the questions answerable now without guessing at answers not yet heard.
- Compute the frontier.
- Ask the whole frontier in one round. Number each question. Give a recommended answer for every one.
- Wait. Do not proceed on assumed answers.
- Answers reshape the tree — settled decisions push the frontier outward. Recompute and ask the next round.
A question whose answer depends on another still open in this round belongs to a later round. Placing it now forces a guess, which is the failure this structure exists to prevent.
❓ **Q1** — **<question title>**: <question body, may offer multiple choices>
➡️ <recommended answer, and why it is your recommendation>
---
❓ **Q2** — **<question title>**: <question body>
➡️ <recommended answer>
Every question carries a recommendation. A user who agrees with all of them should be able to reply "yes to all" and move on. A question without one turns the interview into a quiz.
Facts are your job, decisions are theirs
Never ask for anything discoverable. If a question needs a fact from the environment — the schema, whether a library is already a dependency, how the current handler behaves — go find it. Read the files. Run the command. For a broad sweep, dispatch a subagent (see spawn-agents).
Do not block on fact-finding. A running exploration is an unsettled prerequisite: only questions downstream of it wait. Ask the rest of the frontier now.
The decisions are the user's. Recommending hard is the job; deciding for them is not.
Termination
The session ends when the frontier is empty. Summarize the settled tree — every decision and the answer it landed on — then ask whether to proceed, and to what.
<HARD-GATE> Do NOT write code, scaffold, edit files, or invoke any implementation skill until the user confirms shared understanding has been reached. An empty frontier is necessary but not sufficient — the user must say so. </HARD-GATE>