grill-me

name: grill-me description: "Use when stress-testing, reviewing, or pressure-testing an existing plan, design, or decision. Systematically interrogates every decision point — one question at a time — until a shared understanding is reached."

Grill Me

Use when the user has a plan, design, or decision and wants it thoroughly challenged before committing. Goal: surface hidden assumptions, weak points, and unresolved dependencies — not generate alternatives. For generating options, use brainstorming instead.

Reserve for decisions that are expensive to reverse: data model changes, public API contracts, security posture, infrastructure choices. Skip for small or low-stakes features.

Constraints

• One question at a time. Include a recommended answer when one is clear.
• Do not ask for information you can read yourself — inspect the codebase first.
• Navigate the full decision tree — do not stop early.
• If an answer reveals a gap or contradiction, name it directly and ask the follow-up.

Process

1. Read docs/product/vision.md if it exists. Extract Principles and Anti-goals — these are absolute constraints. Any decision that violates a principle is a blocking issue, not a trade-off to weigh.

2. Read the user's description and inspect relevant codebase context. Identify major decision areas: goals, constraints, data flow, failure modes, dependencies, rollout, validation.

3. Map decisions to validate — starting with principle violations if any exist, then:

   • Product/feature: problem definition, success metric, MVP scope, edge cases, rollout, observability, external dependencies
   • Architecture: component boundaries, data model, API contracts, failure modes, scaling, migration path, operational burden
   • Implementation: fit with existing patterns, risk surface, test coverage, reversibility, performance

4. Start with the most load-bearing decision — the one that invalidates the most downstream choices if wrong. Follow each answer's branch before moving on.

   Format each question as:

   [Topic]: [Question]

   Recommended: [recommendation, or "no strong preference"]

5. After covering all material decision points, check: did any answer reveal a new contradiction, unresolved dependency, or principle violation? If yes, re-enter step 4 at those points automatically — do not wait for the user to ask. Repeat until no new issues surface.

6. Summarize: decisions that are solid, assumptions validated, risks or gaps still open, items that need a decision before proceeding.

7. Ask whether to proceed, revise, or dig deeper into any open item. On revision, re-enter at the affected decision point — do not restart the full interrogation.

Done When

All material decision points covered and user approves to proceed. Suggest implementation-plan to sequence the work.