Skip to main content

grilling

Grills the user relentlessly about a plan, decision, or idea until every branch of the design tree is resolved. Use when stress-testing thinking or when told grill me.

ソース情報

リポジトリ
shipshitdev/skills
ソースの最終更新活動
2026年10月5日 17:32
検出された SKILL.md の言語
英語
スター
37
フォーク
3

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

ファイルエクスプローラー
3 ファイル

SKILL.md を表示中

SKILL.md
ソースの指示 · 読み取り専用プレビュー
name
grilling
description
Grills the user relentlessly about a plan, decision, or idea until every branch of the design tree is resolved. Use when stress-testing thinking or when told grill me.
license
MIT
metadata
{"version":"2.2.2","tags":"interview, grilling, discovery, planning, frontier","author":"Ship Shit Dev","source":"https://github.com/mattpocock/skills/blob/main/skills/productivity/grilling/SKILL.md","upstream_repo":"mattpocock/skills","upstream_ref":"main","upstream_commit":"4588b32ecab9","last_synced":"2026-10-05","license":"MIT"}
when_to_use
grill me, interview me
# Grilling Interview the user relentlessly until every branch of the **design tree** is resolved. This is the reusable interview primitive. Orchestrators (`interview`, `shape`) invoke it; they own grounding, artifacts, and routing. ## Contract Inputs: - A plan, decision, idea, or design question - Optional repo or docs context already loaded by the calling skill Outputs: - Settled decisions across the design tree - Remaining open questions only when a fact-finding sub-agent is still running Creates/Modifies: - None. Calling skills persist briefs, glossaries, or issues. External Side Effects: - Read-only lookups to settle facts. No tracker writes, no code edits. Confirmation Required: - Before acting on the settled tree. Stop at a shared understanding; do not implement. Delegates To: - None. Fact-finding may dispatch a read-only exploration sub-agent. ## Design tree Map the topic as a **design tree**: every decision branches into the decisions that hang off it. The **frontier** is every decision whose prerequisites are already settled — the questions that can be asked *now* without guessing at answers not yet heard. Work the tree in **rounds**. Ask the whole frontier in one round. Wait for answers before the next round. ## Round format Number each frontier question. Recommend an answer so the user can accept it in a word. ```markdown **Q1 — <question title>**: <question body, including choices when they exist> Recommended: <recommended answer> --- **Q2 — <question title>**: <question body> Recommended: <recommended answer> ``` Separate the questions of one round with a `---` rule so each reads on its own. A question whose answer depends on another question still open in this round belongs to a later round, not this one. ## Facts vs decisions Finding **facts** is the agent's job. When a frontier question needs a fact from the environment (filesystem, tracker, docs), dispatch a sub-agent to find it. Do not ask the user for anything that can be looked up. Do not block the rest of the frontier on that lookup. A running exploration is an unsettled prerequisite, so only questions downstream of it wait — ask the rest of the frontier now. **Decisions** are the user's. Put each to them and wait. ## Done when The **frontier is empty**: every branch of the design tree visited, nothing left silently assumed. Do not act on the tree until the user confirms a shared understanding.
GitHubで見る