Reinforce longform-story-design for recovery audit and rebuild

[Hanjo92]

Summary

Reinforce skills/longform-story-design so it can handle long-range Korean fiction recovery work that needs both structural audit and reconstruction planning.

Problem

The current skill is strong for greenfield planning, but audit + repair behavior is scattered across intake, risk, stack, and output sections. It also overlaps with series-qa, which makes routing ambiguous when a user brings a damaged serialized draft.

Goal

• Make longform-story-design the primary skill for multi-arc audit + recovery
• Keep series-qa focused on quick diagnosis, narrow-range QA, and report-first audits
• Ensure recovery requests produce reusable planning artifacts, not only issue lists

Scope

• skills/longform-story-design/SKILL.md
• skills/longform-story-design/references/project-intake.md
• skills/longform-story-design/references/planning-stack-selection.md
• skills/longform-story-design/references/longform-workflow.md
• skills/longform-story-design/references/handoff-to-drafting.md
• optionally skills/series-qa/SKILL.md for boundary clarification

Planned changes

1. Add hard routing near the top of the skill
2. Add one canonical decision order: mode -> package -> risk -> stack -> depth -> drafting slice
3. Add first-class recovery modes for long damaged drafts
4. Define minimum fields and completion criteria for recovery artifacts
5. Clarify reference precedence and drafting handoff after recovery

Acceptance criteria

• Damaged-longform requests route cleanly in one top-level flow
• report-only QA and rebuild-needed recovery are clearly separated
• Recovery outputs include at least one reusable reconstruction package
• Drafting handoff states the active repaired slice
• Main skill and references no longer imply audit + repair; they define it explicitly

Design doc

• docs/superpowers/specs/2026-04-14-longform-recovery-audit-design.md