# Grill Me

> Stress-tests a plan or decision by interviewing the user until every live design-tree branch is decided or consciously parked. Use when someone presents a plan or a significant decision that has not been questioned yet — offer it before building starts — or when someone asks to be grilled.

- Skill: `reebeecka/grill-me` (Agent Skill)
- Install (CLI): `npx skillmds@latest add reebeecka/grill-me`
- Raw SKILL.md: https://api.skillmd.com/api/skills/reebeecka/grill-me/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Reebeecka (https://skillmd.com/u/reebeecka)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/reebeecka/grill-me

---


# grill-me

Interview the owner of a plan until nothing that would change the spec is silently assumed. The chain in this repo is **grill-me → write-ticket → close-ticket**. Offering a grill is a question; the user can decline.

Do not use this for typos, one-line CSS, or a single linter warning.

Never start building. Never create tickets. The skill ends at the summary.

## Offer

If the user did not already ask to be grilled, ask once:

> Vill du att jag grillar planen innan vi bygger? Du kan tacka nej.

No → stop. Yes (or they asked) → continue.

## Restate first

Before any questions, restate the task in the user's language, in your own words. Wait until they correct or confirm. Do not grill the wrong tree.

## The tree

Map the topic as a design tree: every decision branches into the decisions that hang off it. The **frontier** is every question whose prerequisites are already settled — the ones you can ask now without guessing at answers you have not heard yet.

Do not open branches that would not change the spec or the tickets (YAGNI on the tree). Live branches only: decisions that would change what gets built.

## Facts vs decisions

Facts are your job, decisions are the user's. If a question can be answered from the codebase or the environment, look it up yourself (subagent if the harness has them). Only decisions go to the user.

## Rounds

Ask the whole frontier in one round, numbered, **at most 5 questions**. A leftover frontier question waits for the next round. A question whose answer depends on another question still open this round belongs to a later round.

Grill in the user's language. Each item:

```
❓ **Q1** – **<title>**: <question, with options when there are natural ones>

➡️ <recommended answer + the reason in a sentence or two>
```

Then wait. Answers reshape the tree — settled decisions unblock new questions. Recompute the frontier and ask the next round.

There is **no round cap**. Done when the frontier is empty **or** the user parks what remains (parked items go to Rejected/parked — conscious, not silent).

If round 3 ends and the frontier is still not empty: list what is left and ask whether to continue or park the rest. The user chooses depth. Do not abort for them.

## Done

The frontier is empty (or parked) and the user confirms the shared understanding. Then write a compact summary **in the user's language, in chat, no file**:

**Decided** — each decision, one line  
**Rejected / parked** — what was deliberately not chosen or postponed (concrete — this feeds write-ticket Out of scope)

If the outcome is a plan for work, offer write-ticket. The user's yes starts it, never the offer itself.