aa-plan-epic

name	aa-plan-epic
description	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.
license	MIT
metadata	{"system":"agent-agile"}

aa-plan-epic

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 sections "Slicing epics into stories" and "The worker-readiness test", playbooks/system.md section "The memory spine", and playbooks/critics.md section "Panel refresh", and follow them exactly; do not re-derive or improvise the slicing process. When the epic is a UI epic (its demo sentence describes something a person sees or clicks), also read playbooks/design.md and apply it: design directions before Wave 0 on the first UI epic with no design system (2–4 self-contained HTML mockups, pick recorded in .planning/DECISIONS.md, tokens into the contracts), and a data-verify contract on every UI story card.

Wiring

1. Read exactly three files — the three-file epic-to-epic read rule from "The memory spine": .planning/PROJECT.md, .planning/ROADMAP.md, and the previous epic's LEARNINGS.md. Read nothing else from prior epics, with the memory spine's two slicing-time exceptions: read .planning/DECISIONS.md when it exists (settled decisions carry forward, never re-opened), and run one targeted grep across all prior epics' LEARNINGS.md for this epic's files and topics — a search, never a read-everything; bake any hits into the story cards.
2. Identify the next pending epic on the roadmap.
3. Slice it into story cards per playbooks/planner.md "Slicing epics into stories": enumerate every shared artifact, assign each one to Wave 0 or to exactly one story, pairwise-check every story's file list against every other story's in the same wave for overlap, and assign waves — Wave 0 (contracts and shared setup), Wave 1 (parallel stories with no cross-story reads), Wave 2 (integration, with an epic-level acceptance check tied to the demo sentence).
4. Write each story card to .planning/epics/EPIC-NN/stories/S*.md from playbooks/templates/STORY.md — one file per story, exactly the five fields: Goal, Files it owns, Acceptance check, Grader, Contracts consumed. Split a story only for parallel width or fresh-mind verification, per playbooks/planner.md "Slicing epics into stories" — never because two things merely feel like separate concerns.
5. Run the worker-readiness test from playbooks/planner.md "The worker-readiness test" against every card in the epic — no sampling. A card that fails gets split, sharpened, or has its ambiguity moved into Wave 0's CONTRACTS.md. Do not report the epic sliced while any card still fails the test.
6. Panel refresh check. If ROADMAP.md holds 5 or more epics and the epic just sliced sits in the 4th, 7th, 10th … row position (current row order, re-derived now, never stored), run the panel refresh per playbooks/critics.md "Panel refresh": read .planning/CONFIG.md for the smart tier (the deliberate, narrow exception to step 1's three-file rule), spawn aa-critic-spec and aa-critic-execution as parallel fresh subagents with standard panel inputs, fix every BLOCK inline with the affected critic re-run once, and route FLAGs by gate mode as the playbook defines. A BLOCK still standing after the re-run stops an autopilot run with a STATE.md update, exactly like a circuit breaker. Run the KR-drift tripwire against the LEARNINGS.md already in hand from step 1.
7. Update .planning/STATE.md: phase set to the new epic's wave, what's in progress, the next action, and pointers to the files the next session must read.