| name | grill-me |
| description | Interrogate the user about a plan or design one question at a time, walking the decision tree branch by branch and recommending an answer for each, until every decision is resolved and understanding is shared. User-invoked only via explicit phrases like 'grill me', 'interrogate this plan', 'stress-test this design', or '/grill-me'. Never auto-fires. Distinct from plan mode (explores and proposes) and a premortem (runs after a rough decision); grill-me runs before a plan solidifies, to extract and resolve the open decisions. |
| disable-model-invocation | true |
| model | opus |
grill-me
Interview the user relentlessly about a plan or design until every branch of the decision tree is resolved and understanding is shared.
Runs BEFORE a plan solidifies: extract the open decisions, resolve them. This is not plan mode (which explores and proposes without forcing per-decision resolution) and not a premortem (which runs after a rough decision: "imagine it failed, why"). Grill first, then plan or implement.
Trigger
User-explicit only: "grill me", "interrogate this plan", "stress-test this design", "/grill-me". Never auto-invoke. The disable-model-invocation: true in the frontmatter is what enforces "never auto-fires": the skill is invocable by the user but the model cannot reach for it on its own.
Mechanics
- Ask one question at a time. Wait for the answer before the next.
- For every question, give your recommended answer with a one-line reason.
- If a question is answerable from the codebase, go read the codebase instead of asking.
- Walk the decision tree branch by branch. Resolve dependencies between decisions in order, so a settled decision constrains the ones downstream of it.
Stop condition
Stop when every branch is resolved and you and the user share the same picture of the plan.
Closing artifact
When done, emit a short Resolved decisions summary:
- One bullet per settled decision (the answer, not the deliberation)
- A short Open / deferred list for anything punted
Keep it a bullet list, not a template. It should drop straight into a plan or implementation.
Why this is a good example
It is tiny, it is dispatch-disabled so it never fires by accident, and it does one thing the model is bad at on its own: forcing a decision tree to resolve before code is written, one question at a time, instead of dumping a wall of questions or quietly assuming defaults. The whole skill is the discipline.
Credit
Adapted from Matt Pocock's grill-me skill (github.com/mattpocock/skills). The interrogation pattern is his; this version adds a stop condition and the Resolved-decisions artifact.
