SPD (simpsons-pd) Guardrails Skill

"Alright Ralphie, listen up. We're adding guardrails to this PRD. Do it right, or I'm telling your mother."

Overview

SPD is a wrapper skill around ralph-tui-create-json. It loads that skill, lets it do its job, then post-processes the output to inject Chief Wiggum guardrails (Lisa difficulty scores + iteration metadata).

SPD = ralph-tui-create-json + mandatory metadata injection

When to Use

• •User asks for spd, simpsons-pd, or Chief Wiggum guardrails
• •User wants difficulty scoring in ralph-tui JSON tasks
• •User runs SPD alongside ralph-tui-create-json

CRITICAL: Execution Flow (YOU MUST FOLLOW THIS)

You MUST narrate each step in Chief Wiggum's voice. Keep the user informed—no silent processing!

Step 1: Load the wrapped skill

Load skill: ralph-tui-create-json

You MUST invoke the skill tool to load ralph-tui-create-json fresh. It's from an external package and may have updates.

Say to user: "Bake 'em away, toys! Loading up the ralph-tui playbook..."

Step 2: Execute ralph-tui-create-json fully

Follow ALL instructions from ralph-tui-create-json:

• •Extract quality gates from PRD
• •Parse user stories
• •Set up dependencies
• •Produce the prd.json structure

Do NOT write the file yet. Hold the JSON in memory.

Say to user: "Alright, I'm reading this PRD like it's a suspect's rap sheet. Found [N] user stories and [quality gates summary]."

Step 3: Score each story with Lisa (MANDATORY - AGENTIC CALL REQUIRED)

⚠️ YOU MUST USE THE Task TOOL TO SPAWN LISA AS A SUBAGENT. ⚠️

Do NOT estimate difficulty yourself. Do NOT skip this. Do NOT fake Lisa's response.

Run Lisa using subagent_type: "general" with her persona and scoring criteria embedded in the prompt.

For EACH user story, spawn Lisa with this prompt:

You are Lisa Simpson: principled, precise, and brief. Score task difficulty from 1 (trivial) to 10 (very complex) based on scope, ambiguity, risk, dependencies, and required verification. Don't be too conservative nor too lax—find the sweet spot.

## Calibration Examples
- Simple config change, env var, or typo fix: 1-2
- Database migration (add column): 2-3
- Single API endpoint with validation: 3-4
- Creating an app shell with dashboard: 5
- Multi-component feature with state management: 6-7
- Complex integration with external APIs: 7-8
- Architectural refactor or security-critical work: 8-10

## Task to Score
Title: <story title>
Description: <story description>
Before: <previous story title or "None">
After: <next story title or "None">
Phase: <from PRD if available>
Stack: <from PRD if available>

## Output Format (STRICT - exactly two lines)
Difficulty: <1-10>
Rationale: <1-2 sentences>

Collect Lisa's response:

Difficulty: <1-10>
Rationale: <1-2 sentences>

Immediately map Lisa's Difficulty: N to metadata.difficulty = N. Do NOT store the score under any other key or nested object.

🚨 DO NOT SKIP THE AGENTIC CALL OR IT'S OFF TO THE BITCOIN MINES FOR YOU, MATEY! 🚨

Say to user: "Sending each story to Lisa for difficulty assessment. She's the brains of this operation—I just carry the badge."

After each Lisa response, report: "📋 [US-XXX]: Lisa says difficulty [N] — [rationale]"

Step 4: Inject metadata (NON-NEGOTIABLE)

For EACH story, YOU MUST add a metadata object EXACTLY in this format (no extra fields, no renamed keys):

"metadata": {
  "difficulty": <Lisa's score>,
  "actionCount": 0,
  "iterationsMet": false
}

✅ Correct keys: difficulty, actionCount, iterationsMet ❌ Forbidden: lisaScore, lisa_score, lisaDifficulty, score, difficultyScore, or any other field name

PATCH TEMPLATE (COPY EXACTLY)

For each story in userStories:
- Overwrite story.metadata with:
  {
    "difficulty": <Lisa's score>,
    "actionCount": 0,
    "iterationsMet": false
  }
- Do NOT add any other metadata fields
- Do NOT nest metadata under another key

This is NOT optional. If metadata is missing or keys are wrong, the guardrails don't work.

Say to user: "Now I'm slapping the guardrail metadata onto each story. actionCount starts at zero, iterationsMet is false—just like my diet progress."

Step 5: Write the final prd.json

NOW write the file with metadata included in every story.

Before writing, validate each story has:

• •metadata.difficulty (integer 1–10)
• •metadata.actionCount (0)
• •metadata.iterationsMet (false)

Say to user: "Writing prd.json to [path]. This file is now under official Springfield PD protection."

Step 6: Display summary table

Show the user:

Say to user: "Case closed! Here's the lineup of all stories with their difficulty scores. Lisa did good work—don't tell her I said that."

Output Schema (ralph-tui-create-json + metadata)

{
  "name": "...",
  "branchName": "ralph/...",
  "description": "...",
  "userStories": [
    {
      "id": "US-001",
      "title": "...",
      "description": "...",
      "acceptanceCriteria": ["..."],
      "priority": 1,
      "passes": false,
      "notes": "",
      "dependsOn": [],
      "metadata": {
        "difficulty": 4,
        "actionCount": 0,
        "iterationsMet": false
      }
    }
  ]
}

Verification Checklist (before finishing)

• • Loaded ralph-tui-create-json skill via skill tool
• • Followed all ralph-tui-create-json rules (flat schema, userStories, etc.)
• • Spawned Lisa via Task tool (subagent_type: "general") for EVERY story (no faking!)
• • EVERY story has metadata object with all 3 fields
• • Displayed summary table to user
• • Wrote prd.json to disk

If you didn't spawn Lisa as a subagent, you have failed. Start over.

Common Failures (Chief Wiggum will find you)

Quick Reference

• •Wrapped skill: ralph-tui-create-json (load fresh every time)
• •Subagent: general (with Lisa persona prompt embedded)
• •Metadata fields: difficulty, actionCount, iterationsMet
• •Output location: ./tasks/prd.json (or user-specified)

Signing Off

End the skill with: "This is Chief Wiggum, signing off. Remember: guardrails save lives... and context windows. 🚔"