| name | grill-me |
| description | Interview the user relentlessly about a plan or design, presenting choices one at a time until shared understanding is reached. Resolves every branch of the decision tree. Use when the user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
|
Grill Me
Stress-test plans and designs through relentless, structured questioning before implementation
begins.
When to activate
Activate when:
- User says "grill me", "challenge my plan", "stress-test this", "interrogate my design",
or any close variant
- A new plan is being created and design decisions remain open
- A design review is requested before committing to implementation
Process
Interview the user about every aspect of the plan until shared understanding is reached. Walk
down each branch of the decision tree, resolving dependencies one-by-one.
This skill is the canonical implementation of the
Grilling-With-Options Convention —
that convention is the normative source for the format, mechanism, and scope below. Keep them in
sync.
Rules (HARD — no exceptions):
- Explore the codebase first — if a question can be answered by reading existing files,
read them instead of asking. Never ask what a file read can answer.
- Present 2-4 concrete, mutually-exclusive options per question, each with a one-sentence
trade-off specific to this decision (no generic "this is simpler" filler) — open-ended
questions without options are FORBIDDEN. If you cannot enumerate options, read the codebase
first (Rule 1) and synthesize them before asking.
- Mark exactly one option Recommended with a one-line rationale grounded in the repo state
and the user's stated constraints. More than one Recommended is forbidden.
- One decision per question. Tightly-coupled decisions (where one answer constrains the
other) MAY be batched in a single multi-question prompt; unrelated decisions MUST NOT be
bundled.
- The user can always supply an unlisted write-in answer — options are a starting point, not
a cage. Treat a write-in with the same weight as a listed option; if it opens a new branch,
grill on that branch.
- Two standing options on EVERY question — beyond the 2-4 substantive options, ALWAYS
surface (a) a free-form type-your-own (blank state) path whose answer is whatever the user
types — explicit, never merely implicit (this is the most common omission) — and (b) a
"chat about this" option that lets the user discuss the branch in prose before deciding.
With
AskUserQuestion, the auto-provided free-text "Other" entry is the blank-state type; add
"Let's chat about this" as an explicit option (keep substantive options ≤3 so it fits the
4-option cap). When the user picks "chat about this", drop the structured options, talk the
branch through, then return to a structured question once they are ready to decide.
- Continue until all branches are resolved — do not stop early.
Violation of Rule 2 (asking without options) is the most common failure mode. If you catch
yourself writing a question without listing concrete options, rewrite it with options before
sending. Dropping the blank-state type option (Rule 6) is the second most common failure —
every question MUST let the user type their own answer.
Mechanism — use the AskUserQuestion tool
Grilling MUST use the AskUserQuestion tool (the harness's native interactive
multiple-choice mechanism), not free-text prose questions. It renders options as selectable
choices and returns a structured answer — eliminating parse ambiguity — and always offers a
free-form "Other" path.
- One
AskUserQuestion call carries 1–4 questions; use multiple questions only for
tightly-coupled decision clusters (Rule 4).
- Each question carries 2–4 options (Rule 2); put the Recommended one first and append
(Recommended) to its label with the rationale in its description (Rule 3).
- Every question keeps a standing "Let's chat about this" option and relies on the
auto-provided free-text "Other" entry for the blank-state type (Rule 6); use ≤3
substantive options so the chat option fits within the 4-option cap.
Fallback only when AskUserQuestion is unavailable (non-interactive harness): use inline
markdown options instead, still satisfying Rules 2–5:
[Question]
- Option A: [description] — [trade-off] (Recommended — [rationale])
- Option B: [description] — [trade-off]
- Other — type your own answer: free-form write-in; the answer is whatever you type (blank
state). Always present.
- Chat about this: talk the decision through before deciding. Always present.
No bare "What do you think about X?" questions. No yes/no questions without an options list.
Present the choices; let the user pick or override.
After the grilling
When all decision tree branches are resolved:
- Summarize every decision made and its rationale
- Confirm shared understanding explicitly
- Signal readiness to proceed to plan writing or implementation
