ralph-wiggum-marketer

Ralph Marketer

A Claude Code Plugin that provides an autonomous AI copywriter for SaaS content marketing.

DeepWiki: Learn more here

Uses the Ralph Wiggum pattern - an iterative AI loop that ships content while you sleep.

Installation

Option 1: Add as Marketplace (Recommended)

# In Claude Code, add the repo as a marketplace:
/plugin marketplace add muratcankoylan/ralph-wiggum-marketer

# Then install the plugin:
/plugin install ralph-wiggum-marketer@muratcankoylan-ralph-wiggum-marketer

Option 2: Test Locally (For Development)

# Clone the repo
git clone https://github.com/muratcankoylan/ralph-wiggum-marketer.git

# Run Claude Code with the plugin directory
claude --plugin-dir ./ralph-wiggum-marketer

Option 3: Interactive Plugin Manager

# Open the plugin manager:
/plugin

# Browse, search, and install from the interactive UI

Quick Start

# 1. Initialize a new content project
/ralph-init

# 2. Check progress anytime
/ralph-status

# 3. Cancel if needed
/ralph-cancel

Commands

How It Works

…

The Ralph Loop

1. Read PRD: Check scripts/ralph/prd.json for tasks
2. Check Progress: Read scripts/ralph/progress.txt for learnings
3. Pick Task: Find highest priority story where passes: false
4. Execute: Complete the task following acceptance criteria
5. Verify: Run tests to ensure quality
6. Commit: Save progress to git
7. Update: Mark task done, log learnings
8. Repeat: Loop until all tasks complete

Each iteration is a fresh context window. Memory persists through files.

Plugin Structure

…

Database Schema

Input Tables (from other agents)

-- Trends from TrendScout
trends (topic, description, source, relevance_score, status)

-- Research from Research Agent
research (title, summary, key_findings, data_points, category, status)

-- Communications from Product/Marketing
communications (type, title, details, key_messages, target_audience, priority, status)

Ralph's Workspace

-- Content planning
content_plan (content_type, title, brief, target_keywords, status)

-- Work in progress
drafts (plan_id, version, content, word_count, feedback)

-- Final content
published (plan_id, final_content, meta_description)

-- Activity tracking
agent_log (action, details, created_at)

Customizing

Add Your Own Content Sources

Edit src/db/seed.js:

// Add a trend
insertTrend.run(
  'Your Trend Topic',
  'Description of the trend',
  'Source',
  85  // relevance score
);

// Add a communication
insertComm.run(
  'product_update',
  'Your Product Launch',
  'Details about what it does',
  JSON.stringify(['Key message 1', 'Key message 2']),
  'Target audience',
  1  // priority
);

Add Your Own Tasks

Edit scripts/ralph/prd.json:

{
  "id": "WRITE-004",
  "title": "Write your custom blog",
  "acceptanceCriteria": [
    "At least 1000 words",
    "Includes 3 data points",
    "Has compelling CTA"
  ],
  "priority": 5,
  "passes": false
}

Sample Tasks

The default PRD includes 12 stories:

1. SETUP-001: Initialize database
2. PLAN-001: Plan product launch blog
3. WRITE-001: Write launch blog draft
4. PLAN-002: Plan thought leadership blog
5. WRITE-002: Write data-driven blog
6. REVIEW-001: Review and improve draft
7. PUBLISH-001: Publish launch blog
8. PLAN-003: Plan case study
9. WRITE-003: Write case study
10. SOCIAL-001: Create social posts
11. NEWSLETTER-001: Draft newsletter
12. METRICS-001: Log final metrics

The Ralph Philosophy

"Ralph is a Bash loop. Memory persists only through git history and text files. Each iteration is a fresh context window."

Key principles:

• Small stories - Must complete in one iteration
• Explicit criteria - No ambiguity
• Fast feedback - Tests every iteration
• Compound learnings - Patterns accumulate
• Persistence wins - Keep iterating

Credits

• Original Ralph concept: @GeoffreyHuntley
• Official Ralph plugin: claude-plugins-official
• Video walkthrough: @mattpocockuk

License

MIT