| name | grilling |
| description | Relentless one-question-at-a-time interview to stress-test a plan, design, or decision before acting on it — maps it as a design tree, works the open frontier each round, dispatches subagents for anything discoverable instead of asking the user. Use when the user wants to be grilled, asks to stress-test an idea/architecture, or before implementing anything non-trivial where assumptions haven't been surfaced yet. |
Grilling
Interview the user until you both share the same understanding of a plan. Never act on a half-settled one.
Design tree
Map the plan as a tree: each decision branches into the decisions hanging off it. The frontier is every decision whose prerequisites are already settled — askable now, without guessing at an answer not yet heard.
Rounds
Ask the whole frontier at once, numbered, each with your recommendation:
❓ Q1 — <title>: <question, options if relevant>
➡️ recommendation: <your answer>
Wait for the answers before the next round. Each answer reshapes the tree — settled decisions push the frontier outward, unblocking what depended on them. A question whose answer depends on another still-open question belongs to a later round, not this one.
Facts vs decisions
Finding facts is the agent's job, never the user's. A frontier question answerable from the codebase or tools: dispatch a subagent to find it, don't ask. Don't block the round on it — only the questions downstream of that fact wait; ask the rest of the frontier now. Decisions are the user's — put each to them, wait.
Done
Frontier empty, every branch visited, nothing silently assumed. Confirm shared understanding before writing a line of code.
Anti-patterns
| Pattern | Why it fails |
|---|
| Serial questions that could've been one round | Wastes turns; user re-answers a moving target |
| Asking the user something a grep would answer | Wastes their time on lookup, not decision |
| Proceeding on "probably what they meant" | Defeats the point of grilling |
| Dumping questions ignoring dependency order | Later ones get answered before their prerequisites are known |
See also
council — critiques finished code/architecture; grilling stress-tests a plan before it's written.check-work — verifies the result after; grilling reduces how much needs fixing.
