# Amadeus Grilling > Read-only grilling interview. Relentlessly interviews you about a plan, design, or file — round by round over the whole design tree, every question carrying a recommended answer — until every branch is settled and you and the agent reach a shared understanding. Facts are looked up in the codebase; only decisions are put to you. Use when you want to stress-test a plan or design before building. Never advances the workflow stage pointer, never emits audit events, never writes files unless you explicitly ask for the summary to be saved. - Skill: `amadeus-dlc/amadeus-grilling` (Agent Skill) - Install (CLI): `npx skillmds@latest add amadeus-dlc/amadeus-grilling` - Raw SKILL.md: https://api.skillmd.com/api/skills/amadeus-dlc/amadeus-grilling/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: amadeus-dlc (https://skillmd.com/u/amadeus-dlc) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/amadeus-dlc/amadeus-grilling --- # Amadeus Grilling ## Purpose Interview the user relentlessly about every aspect of a plan, design, or decision space until agent and user reach a shared understanding — the design tree worked round by round, every question carrying a recommended answer and its rationale. The discipline is defined ONCE, in `{{HARNESS_DIR}}/amadeus-common/protocols/grilling-protocol.md` (§1 the upstream skeleton — design tree, rounds, frontier; §2 the Amadeus overlay — pruning threshold, deferred nodes, circuit breaker; §3 Question Spec Templates). Read that file first and follow it exactly. This skill adds only the standalone-specific rules below — it does not re-define the discipline. ## Classification Read-only. This skill never advances the workflow stage pointer and never emits an audit event. It is safe to run at any point, with or without an active workflow. The workflow-only obligations in the protocol (questions file, `decision`/`answer` audit logging) do NOT apply here — everything happens in the conversation. ## Standalone rules 1. **Subject** comes from the argument: a file path or a topic. If a file path, read it (and whatever it references) as the starting material for the Investigate step. If no argument was given, ask "What should I grill you about?" as the very first question. 2. **Discipline** is the protocol's §1 and §2 in full: the design tree worked in rounds, the whole pruned frontier asked per round with recommended answers and rationale, facts self-researched with only decisions asked, estimates confirmed with a confidence level, and termination when the pruned frontier is empty or the user says "done". Close with an explicitly confirmed agreement summary that lists the deferred nodes and every unresolved material point. 3. **Level** is the second argument and sets the pruning threshold only (protocol §2.2) — never a question budget. `Minimal` / `Standard` / `Comprehensive` prune progressively less and each arms the circuit breaker at 12 / 24 / 36 rendered questions (protocol §2.4). `Free` prunes nothing and has no breaker: every branch of the tree is visited, and the safety valves are the per-round gate and "done". **Default to Free when the user names no level** — single use matches upstream behaviour. Under Free, say so explicitly in the summary's deferred section ("none — Free prunes nothing") rather than omitting it. 4. **Output is terminal-only.** Print the agreement summary (the full question → decision table, the deferred nodes, and the unresolved material points) to the terminal. Write it to a file ONLY when the user explicitly asks for it to be saved to a path (e.g. "save the summary to docs/plan-review.md") — the same explicit-request exception `/amadeus-outcomes-pack` uses. Never write anywhere unprompted. 5. **No enactment.** Do not start implementing the plan under discussion. The skill ends at the confirmed summary; acting on it is a separate, user-initiated step.