Menu
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/majiayu000/claude-skill-registry --skill ralph-wiggumThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
LLM token logprobs and calibration. Per-decision confidence, ECE, Brier, reliability diagrams, low-confidence triage.
Analyze LLM token logprobs and calibration. Use for per-decision confidence, ECE, Brier scores, reliability diagrams, and low-confidence triage.
回顾最近 N 天的 Claude Code 使用记录——扫描原始会话数据，按主题分组汇总"我都做了什么"，并从个人操作系统视角输出模式、风险与增删建议。当用户说 /recap、"看看我这几天做了什么"、"回顾一下我最近的会话"、"这两天我用 claude 干了啥"、"活动回顾" 时使用。
Based on SOC occupation classification
Showing SKILL.md
| name | ralph-wiggum |
| category | workflow |
| version | 1.0.0 |
| description | Autonomous task completion loop with PRD tracking and LLM memory |
| author | Matt Pocock / Jeffrey Huntley (adapted for NodeJS Starter V1) |
| priority | 2 |
| auto-load | false |
| triggers | ["ralph","autonomous","loop","prd","overnight"] |
| requires | ["verification/verification-first.skill.md","workflow/feature-development.skill.md"] |
The Ralph Wiggum technique is a simple but powerful pattern for autonomous AI-driven development:
passes booleanNamed after Ralph Wiggum from The Simpsons - simple but effective.
"Me fail English? That's unpossible!" - Ralph Wiggum
{
"project": "My Project",
"version": "1.0.0",
"user_stories": [
{
"id": "US-001",
"epic": "Authentication",
"title": "User can sign up with email",
| Field | Purpose |
|---|---|
passes | Boolean gate - only true after verification |
priority | Execution order: critical > high > medium > low |
depends_on | Task IDs that must pass first |
acceptance_criteria | Specific requirements to implement |
attempt_count | Tracks failed attempts (for debugging) |
Append-only file with session entries:
# Ralph Wiggum Progress Log
# Project: my-project
# Created: 2026-01-07T10:00:00Z
---
## Session 1: 2026-01-07T10:30:00Z
**Task**: US-001 - User can sign up with email
**Status**: IN_PROGRESS
### Work Done
- Created SignUpForm component at apps/web/src/components/auth/SignUpForm.tsx
- Added zod schema for email validation
### Issues Encountered
- useAuth hook missing return type annotation
### Learnings
- Always add explicit return types to custom hooks
- Email regex pattern: /^[^\s@]+@[^\s@]+\.[^\s@]+$/
### Next Steps
1. Fix useAuth hook return type
2. Add unit tests for SignUpForm
3. Run full verification
---
## Session 2: 2026-01-07T11:00:00Z
**Task**: US-001 - User can sign up with email
**Status**: COMPLETED
### Work Done
- Fixed useAuth return type
- Added comprehensive unit tests
- All verification passed
### Learnings
- Vitest mocking pattern for Supabase auth
### Next Steps
- Move to US-002
ALL must pass before marking passes: true:
pnpm turbo run type-check # TypeScript compilation
pnpm turbo run lint # ESLint + Ruff linting
pnpm turbo run test # Unit tests (Vitest/Pytest)
pnpm turbo run build # Production build
pnpm --filter=web test:e2e # Playwright E2E tests
┌─────────────────────────────────────────┐
│ Start Iteration N │
└─────────────────┬───────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ All tasks passed? │
│ ├─ Yes → Exit (Project Complete!) │
│ └─ No → Continue │
└─────────────────┬───────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Find highest priority unpassed task │
│ (respecting depends_on constraints) │
└─────────────────┬───────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Load context: │
│ - Read prd.json for task details │
│ - Read progress.txt for learnings │
└─────────────────┬───────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Work on task: │
│ - Implement per acceptance criteria │
│ - Add/update tests │
│ - Follow project conventions │
└─────────────────┬───────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Run verification pipeline │
│ type-check → lint → test → build → e2e │
└─────────────────┬───────────────────────┘
│
┌─────────┴─────────┐
│ │
▼ ▼
┌───────────────┐ ┌───────────────┐
│ All PASS │ │ Any FAIL │
└───────┬───────┘ └───────┬───────┘
│ │
▼ ▼
┌───────────────┐ ┌───────────────┐
│ Set passes: │ │ Increment │
│ true in PRD │ │ attempt_count │
│ │ │ │
│ Git commit │ │ Record issues │
└───────┬───────┘ └───────┬───────┘
│ │
└─────────┬─────────┘
│
▼
┌─────────────────────────────────────────┐
│ Append session to progress.txt │
└─────────────────┬───────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Continue to Iteration N+1 │
└─────────────────────────────────────────┘
# Unix/Mac/WSL
./scripts/ralph.sh --init
# Windows PowerShell
.\scripts\ralph.ps1 -Init
# Creates:
# - plans/prd.json (template)
# - plans/progress.txt (empty)
# - plans/ralph-prompt.md (iteration prompt)
# Unix/Mac/WSL
./scripts/ralph.sh 50 # Max 50 iterations
# Windows PowerShell
.\scripts\ralph.ps1 -MaxIterations 50
# Claude Code command
/ralph run 50
Keep user stories to 1-2 hour scope:
// Good
{ "title": "Add email validation to sign-up form" }
// Too big
{ "title": "Implement entire authentication system" }
Vague criteria lead to incomplete implementations:
// Good
{
"acceptance_criteria": [
"Email format validated with regex",
"Error message shown for invalid email",
"Submit button disabled until valid"
]
}
// Too vague
{
"acceptance_criteria": [
"Form works properly"
]
}
Order tasks logically:
{
"id": "US-002",
"title": "User can log in",
"depends_on": ["US-001"] // Sign-up must work first
}
Check what was learned in previous iterations:
### Learnings
- Vitest mocking pattern for Supabase: vi.mock('@supabase/ssr')
- Always await signUp() before checking error state
Creates checkpoints for recovery:
git log --oneline
# feat(US-003): User can reset password
# feat(US-002): User can log in
# feat(US-001): User can sign up with email
If attempt_count >= 3:
Ralph follows the same patterns as apps/backend/src/agents/long_running/:
| Ralph | Long-Running |
|---|---|
progress.txt | ProgressFile class |
prd.json | FeatureManager |
| Session entries | SessionProgress |
The orchestrator can hand off to Ralph for autonomous completion:
async def handle_large_feature(feature_description: str):
# Create PRD from feature description
prd = generate_prd(feature_description)
save_prd("plans/prd.json", prd)
# Delegate to Ralph
return await delegate_to_agent("ralph-wiggum", {
"prd_path": "plans/prd.json",
"max_iterations": 50
})
Uses the same verification-first principles:
============================================
Ralph Wiggum Technique
Autonomous Task Completion Loop
============================================
>>> Checking prerequisites...
Claude CLI found
PRD file found
All prerequisites OK
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Iteration 1: US-001
User can sign up with email
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
>>> Invoking Claude Code...
[Claude implements sign-up form]
>>> Running verification pipeline...
Running type check...
Type check: PASS
Running lint...
Lint: PASS
Running tests...
Tests: PASS
Running build...
Build: PASS
Running E2E tests...
E2E: PASS
Verification passed! Marking US-001 as complete.
[Auto-commit: feat(US-001): User can sign up with email]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Iteration 2: US-002
User can log in with email
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
...
============================================
Summary
============================================
Iterations: 5
Tasks Completed: 5
Progress: 5/5 tasks complete
Duration: 00:45:23
All tasks complete! Project finished in 5 iterations.
All implementations follow Australian defaults: