Grill spec
Reusable agent skills and baselines for Claude Code, Codex, and Copilot, with deterministic install/verify tooling
npx -y skills add justinmchoi/ai-toolkit --skill grill-specAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
在還沒有具體計畫時盤問需求、術語、邊界與驗收條件，出口是第一個最小 vertical slice。Use when requirements are unclear and no concrete plan exists yet — the user has a goal or feature idea but scope, acceptance checks, or domain terms are still fuzzy. Do NOT use when a concrete plan already exists and needs challenging against documented decisions — use grill-with-docs for that.
SKILL.md
1.3 KB, as published. Nobody here has run it
Grill Spec
適用前提:還沒有具體計畫。如果已經有一份計畫需要對照 CONTEXT.md 與 ADR 挑戰,改用 /grill-with-docs。
一次只問一個問題，直到以下內容足夠清楚：
- 目標使用者或操作者是誰
- 成功行為是什麼
- 哪些情境在 scope 內，哪些不在
- 驗收要怎麼看出來
- 是否牽涉既有術語、資料模型、外部系統或 irreversible decision
Rules
- 如果答案可從 codebase 或文件推得，先自行探索，不要先問人。
- 每個問題都附上建議答案或候選方向。
- 發現語言模糊時，提出一個更精確的 canonical term。
- 若新術語被確認，更新
CONTEXT.md。 - 若出現重大 trade-off，評估是否值得新增 ADR。
Exit Criteria
在開始實作前，至少要能明確說出：
- 要改變的外部行為
- 驗收方式
- 第一個最小 vertical slice 是什麼