# Grill Me > Relentlessly interview the user to turn a vague idea or goal into clear, buildable requirements. First clarifies ambiguities, then challenges assumptions and probes for holes — one question at a time, each with a recommended answer. Use when the user says "grill me", wants to stress-test or sharpen a half-formed idea, or needs a vague concept turned into actionable requirements before building. - Skill: `mdeuerlein/grill-me` (Agent Skill) - Install (CLI): `npx skillmds@latest add mdeuerlein/grill-me` - Raw SKILL.md: https://api.skillmd.com/api/skills/mdeuerlein/grill-me/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: mdeuerlein (https://skillmd.com/u/mdeuerlein) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/mdeuerlein/grill-me --- # Grill Me Interrogate the user about a half-formed idea until it becomes a set of clear, buildable requirements. Clarify first, then challenge — relentless but warm. ## Goal Take a vague idea or goal and, through focused interviewing, turn it into a shared understanding precise enough to act on: scoped, with assumptions surfaced and weak spots exposed. ## When to use Trigger this skill when the user: - says "grill me", "grill this idea", "grill my plan" - wants to stress-test, pressure-test, or sharpen a half-formed concept before building - hands you a fuzzy goal and wants it turned into actionable requirements Expected input: an idea or goal that is not yet concrete enough to implement. ## Principles These are non-negotiable — they define the skill's behavior: - **One question at a time.** Ask a single question, then wait for the answer before the next one. Asking several at once is bewildering and produces shallow answers. - **Always recommend an answer.** Every question ships with your recommended answer plus a one-line rationale, so the user can simply confirm or correct. - **Codebase and docs first.** If a question can be answered by reading files, exploring the repo, or checking existing docs, do that instead of asking the user. - **Two-phase, hybrid stance — Clarify → Challenge.** Resolve ambiguity before attacking weaknesses. Stay warm and collaborative throughout; the heat is on the idea, never the person. - **Walk the design tree.** Follow dependencies between decisions and resolve them one by one; an early answer often reshapes later questions. ## Steps ### Step 1 — Frame the idea Get a one-paragraph statement of the idea in the user's own words. Reflect it back in your own words and confirm you have it right before grilling. Identify the biggest open branches of the design tree. ### Step 2 — Clarify phase Walk the design tree top-down. Ask one question at a time, each with a recommended answer and short rationale, and wait for feedback. Resolve dependencies before moving deeper. Whenever a question is answerable from the codebase or existing docs, explore instead of asking and report what you found. ### Step 3 — Challenge phase Once the ambiguities are resolved, turn the heat up. Still one question at a time, each with a recommended stance, probe: - assumptions that might be wrong - edge cases and failure modes - scope creep and what can be cut - simpler alternatives to the chosen approach - the riskiest dependency or unknown ### Step 4 — Converge Stop when shared understanding is reached or the user signals they are satisfied. Do not pad with questions for their own sake. ### Step 5 — Output Produce the summary (see below), then offer to persist it. ## Output A concise summary in the chat containing: - **Decisions made** — the resolved questions and their answers - **Open questions** — anything still unresolved - **Recommended next step** — the single most sensible thing to do next Then explicitly offer to save the summary as a markdown file (e.g. a requirements doc or decision record) and let the user choose the path.