Grill Me — Stress-Test an Idea

Interview me relentlessly about every aspect of this plan until we reach a shared understanding.

Rounds

Model the open questions as a dependency graph: a question is blocked when its best framing depends on an answer we don't have yet, unblocked otherwise. Then, each round:

1. Ask only the questions unblocked right now — never one whose framing depends on an unanswered question.
2. Batch them into a numbered round (Q1, Q2, …). For each question, provide your recommended answer — don't just ask, bring an opinion.
3. When the answers land, push the frontier: newly unblocked questions form the next round.

I can answer by number — "Q1 agree, Q2 agree, Q3 change to X" — so a round of easy questions costs one turn, not five.

The graph rule is what makes batching safe: batching without it asks dependent questions out of order, which is the problem asking one-at-a-time was avoiding.

If a question can be answered by exploring the codebase, explore the codebase instead of asking.

If a question can be answered by searching the knowledge base, search it — prior work may already have answers.

When to stop

Stop when every branch of the decision tree is either:

• Resolved — we agreed on an answer
• Explicitly deferred — we agreed it's out of scope or future work
• Flagged as a risk — we identified it but can't resolve it now

Summarize what was resolved, what was deferred, and what risks remain.

Connection to workshops

If the grilling reveals enough structure and the user wants to proceed, suggest transitioning to /spec (depth auto-selects) to formalize it. The grilling output becomes input to the problem statement.

When the effort is too foggy and too large for one interview, run /chart first — it schedules grill-me sessions rather than replacing them.

Inspired by Matt Pocock's grill-me skill; dependency-graph rounds from his v1.2. Adapted with KB integration and workshop pipeline connection.
Source: github.com/mattpocock/skills/tree/main/grill-me