Requirement Planning Skill

Overview

This skill produces a complete, scored execution plan from a user's requirement input. It runs in 6 sequential phases. Each phase has a dedicated reference file with exact prompt instructions and quality criteria.

Do not skip phases. Do not merge phases. Each phase has a distinct role.

---

Phase Map

User Input (requirement background + dev points + edge cases)
          │
          ▼
  [Phase 1] Project Discovery      → references/01-project-discovery.md
          │
          ▼
  [Phase 2] Technical Plan         → references/02-technical-plan.md
          │
          ▼
  [Phase 2b] TODO Checklist        → references/02b-todo-checklist.md
          │
          ▼
  [Phase 2c] Task Persistence      → references/02c-task-persistence.md
          │
          ▼
  [Phase 3] Edge Case Analysis     → references/03-edgecase-analysis.md
          │
          ▼
  [Phase 4] Plan Validation        → references/04-plan-validation.md
          │
          ▼
  [Phase 5] Test Cases             → references/05-test-case.md
          │
          ▼
  [Phase 6] Scoring Summary        → references/06-scoring-summary.md
          │
          ├─ Score ≥ 8 → ✅ Deliver final plan
          └─ Score < 8 → 🔄 Return to Phase 2, revise plan

---

How to Execute This Skill

Step 1 — Read all reference files first

Before responding to the user, read all 6 reference files:

references/01-project-discovery.md
references/02-technical-plan.md
references/02b-todo-checklist.md
references/02c-task-persistence.md
references/03-edge-case-analysis.md
references/04-plan-validation.md
references/05-test-cases.md
references/06-scoring-summary.md

This gives you the full prompt template for each phase before you start.

Step 2 — Check what the user has already provided

Scan the user's input for:

• Requirement background (context, goal, why)
• Development points (what needs to be built)
• Edge cases (known boundary scenarios)
• Entry files / core files (if any)
• Expected completion state (if any)

If all five are present → skip Phase 1 questions, proceed to Phase 2. If any are missing → run Phase 1 to collect them.

Step 3 — Execute phases in order

Run each phase fully before starting the next. Output each phase with a clear header:

## Phase 1: Project Discovery
...

## Phase 2: Technical Plan
...

Step 4 — Apply the re-plan rule at Phase 6

If the Phase 6 score is < 8/10 → do NOT deliver the plan. Instead, explain which dimensions failed, revise Phase 2 plan accordingly, and re-run Phases 3 → 4 → 5 → 6.

---

Output Format

Each phase outputs a structured block. The final deliverable is:

1. Phase 2 plan (revised if needed)
2. Phase 2b TODO checklist (IMPL / VERIFY / TEST × NORMAL / EDGE / RARE)
3. Phase 2c

   tasks.json

   — machine-executable task store, one row per TODO
4. Phase 3 edge cases table
5. Phase 5 test cases (unit tests with 100% coverage target)
6. Phase 6 score card

tasks.json

is the primary execution artifact. It is consumed task-by-task during implementation. Each task's

query

field is a self-contained instruction requiring no additional context to execute.

---

Quality Invariants

• Every plan must reference actual files from the codebase (from Phase 1 discovery)
• Every edge case must map to at least one test case in Phase 5
• Every test case must reference a specific function or widget
• The plan score must be computed honestly — never inflate to avoid re-planning