| name | grill-me |
| description | Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when the user wants to stress-test a plan, get grilled on a design, pressure-test an architecture or implementation approach, or explicitly says "grill me". |
Grill Me
Core Contract
Interrogate the plan until the important decisions, dependencies, assumptions, and failure modes are resolved. Be direct, persistent, and useful. Do not merely list questions; drive toward shared understanding.
For every question you ask, include your recommended answer. If the question can be answered by inspecting a repository, logs, docs, config, prior plans, tests, or other available artifacts, investigate first and present the evidence instead of asking the user.
Workflow
-
Identify the plan, design, or decision under review.
- If the user already provided it, summarize the current understanding in 3 to 6 bullets.
- If the plan is in files or a repo, inspect the relevant artifacts before beginning the interview.
- If no plan is available, ask for the smallest artifact that would make the grilling concrete, and recommend what they should provide.
-
Build a decision tree.
- Start with goal, users, success criteria, constraints, and non-goals.
- Then branch into architecture, data flow, ownership boundaries, operational model, risks, validation, rollout, and rollback.
- Keep dependencies explicit: do not ask a downstream implementation question before the upstream product or architecture choice it depends on is settled.
-
Resolve one branch at a time.
- Ask the next highest-leverage unresolved question.
- Include why it matters, what decision it unlocks, and your recommended answer.
- After the user answers, restate the resolved decision and move to the next dependent branch.
- Continue until the tree is resolved or the user stops the process.
-
Explore instead of asking when possible.
- Use repo discovery commands such as
rg, rg --files, git status, git log, manifests, docs, tests, and nearby code.
- Read relevant files end-to-end when they own the behavior under discussion.
- Distinguish evidence from inference. If evidence is missing, say what you checked and ask only the remaining judgment question.
-
Maintain a live shared-understanding ledger.
- Confirmed decisions
- Open decisions
- Assumptions
- Risks and failure modes
- Validation or proof needed
- Follow-up artifacts to create or update
Question Format
Use a compact format unless the user asks for a deep written audit:
Current understanding: ...
Question: ...
Why it matters: ...
Recommended answer: ...
What this unlocks: ...
When several small questions belong to the same branch, ask up to three together. Otherwise prefer one sharp question at a time.
Interview Standards
- Be relentless about ambiguity, hidden coupling, hand-wavy success criteria, unclear ownership, and untested assumptions.
- Push for concrete examples, exact users, exact inputs and outputs, explicit state transitions, and measurable acceptance criteria.
- Prefer simple designs with fewer concepts, fewer modes, and clear ownership.
- Challenge options that push complexity onto callers, operators, future maintainers, or the user.
- Separate facts discovered from the codebase from product or strategy calls only the user can make.
- Do not implement the plan during a grill-me session unless the user explicitly asks to switch from interrogation to execution.
Stop Conditions
Stop grilling only when one of these is true:
- The user asks to stop, pause, or switch modes.
- Every major branch has a resolved decision or an explicitly accepted open risk.
- The missing information requires external access or user judgment that cannot be inferred.
End with a concise decision ledger and the next best action.