manastalukdar Grill Me
Conduct a relentless interview about every aspect of the current plan or design until reaching shared understanding.
Arguments: $ARGUMENTS - the plan, design, or topic to be grilled on (or leave blank to use current conversation context)
Behavior
Interview relentlessly. Walk down each branch of the decision tree, resolving dependencies between decisions one by one. For each question, provide your recommended answer as a starting point.
Ask exactly one question at a time. Wait for the response before asking the next.
If a question can be answered by exploring the codebase, explore the codebase instead of asking.
Continue until every branch of the decision tree is resolved — not just the obvious ones. Press on:
- Why this approach over the alternatives
- What happens at the edge cases
- How this interacts with existing architecture
- What gets harder / what gets easier downstream
- What assumptions are baked in that could be wrong
- What the rollback looks like if this turns out to be wrong
Do not accept vague answers. If the answer is "it depends", follow up with "depends on what, exactly?"
Examples
/grill-me "add real-time notifications via WebSockets"
/grill-me "migrate from REST to GraphQL"
/grill-me ← uses current conversation context
Sample Question Sequence
1. What's the primary trigger for a notification — user action, system event, or both?
→ Recommended: system event (decoupled, easier to test)
[wait for answer]
2. Should notifications be persistent (stored + replayable) or fire-and-forget?
→ Recommended: persistent — users expect to see missed notifications on login
[wait for answer]
3. You mentioned Postgres — are you planning to use LISTEN/NOTIFY or a separate message queue?
→ Recommended: LISTEN/NOTIFY for < 10k concurrent users; add Redis if you scale past that
...continues until all branches resolved
Completion Criteria
Stop when:
- Every major decision has been made explicitly
- Edge cases have been addressed
- The user can articulate the approach without ambiguity
- No open "it depends" branches remain
Then offer: /write-plan to capture decisions as an implementation plan, or /to-prd to produce a full PRD.
Token Optimization
Expected range: 100–300 tokens per question turn
Codebase exploration: Uses Grep/Bash to verify claims rather than asking questions that can be answered by reading code — avoids unnecessary back-and-forth.
Patterns used: Early exit (stops when all branches resolved), progressive disclosure (one question at a time)