aa-panel

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"}

aa-panel

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.

Wiring

1. Read the draft plan: .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).
2. Read .planning/CONFIG.md for the smart_tier and commercial fields.
3. Spawn 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.
4. Collect every subagent's Verdict format block.
5. Fix every 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.
6. Present the remaining FLAG findings plus the fixed plan to the user; the user's decision is the gate.
7. If the market critic's verdict is 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.
8. Update .planning/STATE.md to record the outcome of the gate.