菜单
Use when the user has a rough or vague product idea that needs interrogation before any planning — or invokes /aa-grill. Locks the idea into .planning/IDEA.md or recommends against building it.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Use when the user has an existing plan, PRD, or spec document they want brought into this planning system instead of a fresh grill and intake — or invokes /aa-import with a path or description of that document.
Use when the user wants to plan a new project from a clear idea, has already locked an idea with aa-grill, or invokes /aa-new-project.
Use when multiple epics should run back to back without a human kicking off each one by hand, or the user invokes /aa-autopilot [--gate full-auto|checkpoint].
Use when the current epic on the roadmap is done and the next pending epic needs to be sliced into worker-ready story cards, or the user invokes /aa-plan-epic.
Use when the user wants to review a finished epic's demo, decide whether to approve, redo, or replan an epic, or invokes the review gate, for example "/aa-review", "let's review this epic", "walk me through the demo", "I want to give feedback on this epic".
Use when a sliced epic (story cards already written) is ready to build, or the user invokes /aa-execute-epic.
基于 SOC 职业分类
| name | aa-grill |
| description | Use when the user has a rough or vague product idea that needs interrogation before any planning — or invokes /aa-grill. Locks the idea into .planning/IDEA.md or recommends against building it. |
| license | MIT |
| metadata | {"system":"agent-agile"} |
Resolve the Agent-Agile playbook root: use the first of these that exists — (1) ${CLAUDE_PLUGIN_ROOT}/playbooks, (2) ./.claude/agent-agile/playbooks, (3) ./.agents/agent-agile/playbooks, (4) ~/.claude/agent-agile/playbooks, (5) ~/.agents/agent-agile/playbooks, (6) ./playbooks.
Read playbooks/planner.md section The Grill in full, including the Grill → intake mapping table inside it, and follow it exactly — do NOT re-derive or improvise the system. The interrogation script, its handling rules, and the valid "recommend not building it" outcome are all defined there; use them as written.
This is a live conversation, not a subagent job — it runs in the main loop. Ask the seven grill questions from the playbook one at a time, in order. Never dump the list at once. Apply the playbook's handling rules on every answer: a vague answer gets narrowed into a forced choice built from the answer's own words before moving on; "everyone" and "it'd be cool" are rejected outright per the playbook.
If the idea fails its own questions — no real user, no real switch reason, no number that would move — say so plainly, name which questions it failed, and recommend against building it. Stop there; do not write IDEA.md for an idea you've just recommended against.
Once all seven questions produce real answers:
.planning/IDEA.md from the template at playbooks/templates/IDEA.md, filling Problem, Exact user, The one action, Why now / why switch, Non-goals, and Success signal from the grill answers. Once written, IDEA.md is locked — do not reinterpret it later..planning/STATE.md from the template at playbooks/templates/STATE.md: set Project to a short working name for the idea (confirm it with the user in one breath before writing it), Phase to grill, Now to "grill complete, IDEA.md locked", Next to "run /aa-new-project", Pointers to .planning/IDEA.md, and Blockers to none./aa-new-project as the next step. State, using the Grill → intake mapping table, exactly which of the 8 intake questions are already pre-filled from this conversation and which three (hard constraints, riskiest unknown, review-batch appetite) will still be asked fresh.