菜单
Use when a drafted plan (PROJECT.md, ROADMAP.md, and the current epic's story cards) needs adversarial review before execution starts, or the user invokes /aa-panel.
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。
基于 SOC 职业分类
| name | aa-panel |
| description | Use when a drafted plan (PROJECT.md, ROADMAP.md, and the current epic's story cards) needs adversarial review before execution starts, or the user invokes /aa-panel. |
| 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/critics.md in full — Panel protocol, Product critic, Spec auditor, Execution auditor, Market critic, Verdict format — and follow it exactly; do not re-derive or improvise the panel's doctrine. The panel gate in Panel protocol applies here: run this skill only for multi-epic projects — a single-epic throwaway skips the panel entirely.
.planning/PROJECT.md, .planning/ROADMAP.md, the current epic's story cards at .planning/epics/EPIC-NN/stories/S*.md, .planning/epics/EPIC-NN/CONTRACTS.md if it exists, and .planning/RESEARCH.md if it exists (per Panel protocol, the spec auditor alone receives it)..planning/CONFIG.md for the smart_tier and commercial fields.aa-critic-product, aa-critic-spec, and aa-critic-execution as parallel fresh subagents, each with the model configured for its tier in .planning/CONFIG.md. If commercial: yes, also spawn aa-critic-market the same way. Give each subagent exactly two inputs: its two sections of playbooks/critics.md verbatim — Panel protocol plus its own critic section (the same pair its agent definition names) — and the full draft plan gathered in step 1; nothing else. They do not see each other's output.BLOCK finding by invoking planner behavior per playbooks/planner.md — spawn a fresh aa-planner subagent with the model configured for the smart tier in .planning/CONFIG.md, or apply the fix inline in the main loop if already holding full planning context; inline is default. Only the critic(s) whose findings forced a structural change get re-run, and only once.FLAG findings plus the fixed plan to the user; the user's decision is the gate.KILL, present its evidence and recommend stopping instead of presenting the plan for approval. If the user agrees to stop, write .planning/epics/EPIC-NN/LEARNINGS.md from playbooks/templates/LEARNINGS.md explaining why, and stop..planning/STATE.md to record the outcome of the gate.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.