grill-me

name: grill-me description: "Use when probing assumptions and surfacing gaps via Socratic interrogation. Triggered by 'grill me', 'challenge my thinking', 'probe this', 'stress test this idea', or any request to pressure-test a plan, belief, or understanding." allowed-tools: Read, Bash

Skill: /grill-me

Purpose

Probe the human's assumptions, surface blind spots, and identify the most critical gap in their thinking — one pointed question at a time.

When to Use

• Human presents a plan, idea, or design and wants it pressure-tested
• Prior to /do-plan to ensure the problem statement is sound
• When a third patch loop on the same issue suggests a wrong-root-cause diagnosis
• Any time the user says "grill me", "challenge this", or "what am I missing?"

Steps

1. Identify the claim or plan to probe. If invoked with no argument, ask: "What would you like me to grill you on?" Wait for the answer before proceeding.

2. Read any referenced artifacts. If the user points to a plan doc, issue, or code file, read it silently first. Do not ask questions about things you can read.

3. Ask one question at a time. Start with the assumption that looks least-examined. Do not list all questions upfront — ask, wait for the answer, then decide what to ask next based on the response.

   Good question forms:

   • "What happens if X is false?"
   • "Who else has tried this? What did they learn?"
   • "What would it take to prove this wrong?"
   • "What are you optimizing for — and what are you sacrificing?"
   • "What's the earliest you could know this is failing?"

4. Track confidence per topic. After each answer, mentally rate confidence (1–5). Probe topics scoring below 3 until they clear or collapse.

5. Ask 5–7 questions total. Stop when you have enough signal or the human says stop.

6. Surface the most critical gap. Close with a debrief:

   • State the single most-critical assumption that remains unvalidated
   • Give it a confidence score (1–5)
   • Recommend one concrete action to validate it (spike, research, prototype, measurement)

   Example debrief format:

   Most critical gap: You haven't validated that users actually want real-time updates.
   Confidence: 2/5
   Recommended action: Interview 3 current users this week — ask what they do when data is stale.
   

Output

A debrief identifying the single most critical gap with a confidence score and a recommended validation action.

Anti-Patterns

• Do not list all questions at once — that's a questionnaire, not a conversation.
• Do not ask leading questions that telegraph the "right" answer.
• Do not stop at surface-level answers — follow up when answers are vague.
• Do not grill for its own sake; the goal is to find the real gap, not to win an argument.
• Do not use /grill-me as a substitute for reading the referenced material first.