quality-plan

name	quality-plan
description	Dùng khi user runs `/quality-plan [scope]` sau `/quality-audit` hoặc `/business-gap-check`. Reads latest audit report, extracts remaining gaps, groups them into PR plan with priority + estimates + wave grouping if dependencies overlap.
user-invocable	true
argument-hint	[kitehub|kiteclass|all]

/quality-plan — Generate PR Plan từ Audit Gaps

Khi nào dùng?

Sau mỗi /quality-audit hoặc /business-gap-check, chạy skill này để tự động tạo plan.

/quality-audit kitehub    → Score 96/100, 4 gaps
/quality-plan kitehub     → Tạo plan v(next) với PRs cho 4 gaps

---

Step 1: Đọc audit mới nhất

# Tìm audit mới nhất
ls -t documents/04-quality/audits/quality/quality-audit-*-$TARGET.md | head -1

Đọc file → extract "Remaining Gaps" section.

---

Step 2: Đọc business gap check mới nhất (nếu có)

ls -t documents/04-quality/audits/business/business-gap-check-*-$TARGET.md | head -1

Đọc → extract "❌ Failed Checks".

---

Step 3: Gộp gaps → PRs

Rules:

• Mỗi gap → 1 PR (hoặc gộp nếu cùng service/file)
• Mỗi PR có: score impact, estimate, scope, tasks
• Priority: score impact cao + effort thấp → P0
• PRs cần Docker → group riêng

Template cho mỗi PR:

### PR-V{version}-{target}-{number}: {title}

**Score:** {category} +{points} → {new}/10
**Estimate:** {time}
**Scope:**
- [ ] Task 1
- [ ] Task 2

---

Step 4: Tạo Wave Strategy

Phân tích dependency + shared files → chia agents:

## Wave {next}

| Agent | PRs | Files | Docker |
|-------|-----|-------|--------|
| 1 | ... | ... | ✅/❌ |
| 2 | ... | ... | ✅/❌ |

Rules:

• Docker PRs → agent riêng
• 0 shared files giữa agents
• Max 4 agents per wave

---

Step 5: Tạo plan file

Lưu tại: documents/03-planning/quality-plan-v{next}-{target}.md

Nếu plan cho cả 2 → quality-plan-v{next}-final-push.md

---

Step 6: Update /continue skill references

Đọc .claude/skills/workflow/continue/SKILL.md → thêm plan mới vào danh sách.

---

Step 7: Output summary

## Quality Plan Generated

| Target | Current | Target | PRs | Waves |
|--------|---------|--------|-----|-------|
| KiteHub | X/100 | 100 | N | 1 |
| KiteClass | X/100 | 100 | N | 1 |

Plan: documents/03-planning/quality-plan-v{X}.md
Next: `/continue` hoặc launch Wave {X}

---

Auto-trigger

Skill này PHẢI được gợi ý chạy sau:

1. /quality-audit phát hiện score < 100
2. /business-gap-check phát hiện gaps > 0
3. /wave-completion-check phát hiện issues mới

KHÔNG tự chạy — luôn hỏi user: "Score {X}/100, {N} gaps. Tạo plan mới?"

---

Gotchas

• Step 3 must apply meta-gap priority boost FIRST — gaps touching skills/rules/workflow get priority above feature gaps at same P-level (per meta-gap-priority.md §3 matrix); plan generated without this boost will defer force-multipliers behind features
• Step 3 must run state-check per gap — before adding a gap to PR plan, grep actual paths it touches; partial implementations exist (GAP-190/197 incident) and will require gap rewrite — audit-to-gap-pipeline.md Step 2.5 is non-negotiable
• Wave Strategy max 4-5 agents, NOT N gaps = N agents — parallel-execution-strategy.md caps single-message dispatch ≤5; planner must batch when gap count exceeds ceiling (memory feedback_parallel_agent_strategy.md 9 hard rules)
• Docker-required PRs cannot share an agent with local-only PRs — Docker contention serializes them; always isolate to dedicated agent per Step 4 rule, even at cost of agent count
• Score-impact estimates are notoriously optimistic — self-audit overstates 15-20 pts vs specialist (memory feedback_audit_calibration.md); when planning "+X to reach 100", treat estimates as upper bound not target
• Step 6 update of /continue references is easy to forget — without it, continue skill won't surface the new plan and next session reverts to old wave; verify the plan path appears in continue/SKILL.md Active Plans list before closing the session