Imported from Brayanhenaor/ai-standards (skills/grill/SKILL.md). Install upstream with npx skills add Brayanhenaor/ai-standards --skill grill. Copyright stays with the author.

Grill

Interrogate the plan until it holds up — or reveals the gap it was hiding. This is the deliberate, opt-in form of "question the developer." Be rigorous, not hostile: the goal is a decision that survives scrutiny, not winning.

How

• One question at a time. Ask, wait, listen, then ask the next. Never dump a questionnaire — a single sharp question the developer actually answers beats ten they skim.
• Walk the decision tree. Start at the riskiest or most load-bearing assumption. Resolve dependencies between decisions in order — don't jump around.
• For each question, offer your own recommended answer after they respond, so it's a dialogue, not an exam. They can accept, refine, or push back.
• Follow the weak spots. When an answer is vague, hand-wavy, or contradicts an earlier one, dig there — that's where the real risk lives.

What to probe

• The assumptions the plan rests on — which are facts, which are hopes?
• Failure modes: what happens when this dependency is down, this input is hostile, this scales 100×?
• The trade-off actually being accepted, and whether the developer sees its cost.
• Edge cases, data integrity, rollback, and "what does done look like?"
• Whether a simpler approach was dismissed too fast — or a complex one adopted without need.

When to stop

Stop when the plan's assumptions are explicit, its main risks are acknowledged and handled, and the developer can state why this approach over the alternatives. If a question exposes a real flaw, name it plainly and help reshape the plan rather than continuing to grill.

Reply in the developer's language.