# Grill > Relentless multi-round interview that resolves every open decision in a plan or design before building starts. Use when the user says "grill me", wants their thinking stress-tested, wants to align before implementing, or is about to start a build from a vague brief. Distinct from loose ideation and from a single clarifying question. - Skill: `presidenteog/grill` (Agent Skill) - Install (CLI): `npx skillmds@latest add presidenteog/grill` - Raw SKILL.md: https://api.skillmd.com/api/skills/presidenteog/grill/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: PresidenteOG (https://skillmd.com/u/presidenteog) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/presidenteog/grill --- # Grill Adapted from https://github.com/mattpocock/skills (mattpocock/skills, MIT) — rewritten, not copied. ## Overview The most common failure in a build is misalignment found after the code exists. The fix is to spend the questions up front: interview the user until every branch of the design is decided, not assumed. Model the design as a **tree** — each decision spawns the decisions that hang off it — and work it breadth-first in rounds. ## When to use - Before implementing anything from a brief that still has open choices. - When the user asks to have a plan, spec, or idea stress-tested. - Not for open "what could we build" ideation: that is `superpowers:brainstorming`. Not for a scored objective review with a dissenter: that is `/prompt-scan`. Not for one missing fact: that is a single `AskUserQuestion`. ## The loop 1. **Map the frontier.** The frontier is every decision whose prerequisites are already settled: the questions you can ask now without guessing at answers you have not heard. A question whose answer depends on another still-open question belongs to a later round. 2. **Ask the whole frontier in one round.** Number each question, give your recommended answer. 3. **Wait for answers.** Do not proceed on assumptions. 4. **Recompute.** Each answer settles decisions and pushes the frontier outward, unblocking downstream questions. Return to step 1. 5. **Done when the frontier is empty**: every branch visited, nothing silently assumed. Confirm shared understanding before acting. ## Rules - **Facts are your job, never the user's.** When a frontier question needs a fact from the environment (a file, a dependency version, an API shape), dispatch a subagent (`cavecrew-investigator`, or `Explore`) to find it. Do not ask the user what you can look up. - **Do not block on fact-finding.** A running exploration is an unsettled prerequisite: only its downstream questions wait. Ask the rest of the frontier now. - **Decisions are the user's.** Put each to them and wait. - Prefer `AskUserQuestion` when the round has 4 or fewer questions and each has discrete options. Otherwise use the numbered format below. ## Round format (when not using AskUserQuestion) ``` Q1 - : Recommended: --- Q2 - : Recommended: ``` ## This workspace Personal Context rule "large task -> brief plan first, wait for approval" still holds: a grilling session resolves the design, then the plan goes to the user before code. If the work will outlive one session, the resolved design is written out via `persistent-file-planner`.