---
name: ralph-leaks
description: "Auto-loads all system prompt leaks from major AI providers (Anthropic, OpenAI, Google, Perplexity, xAI, Proton) for Ralph's autonomous agent selection"
---

# Ralph Leaks Integration

This directory contains **91 skills and 88 agents** converted from the [system_prompts_leaks](https://github.com/asgeirtj/system_prompts_leaks) repository. These are actual system prompts from major AI providers.

## Auto-Integration with Ralph

All these skills and agents are automatically available to Ralph's autonomous agent selection system.

## Available System Prompts

### Anthropic Claude (21 skills + 21 agents)
- `anthropic-claude-code` - Claude Code CLI system prompt
- `anthropic-claude-4-5-sonnet` - Claude 4.5 Sonnet full system prompt
- `anthropic-claude-sonnet-4` - Claude Sonnet 4 complete system message
- `anthropic-claude-opus-4-5` - Claude Opus 4.5 thinking prompt
- `anthropic-claude-in-chrome` - Claude for Chrome extension
- `anthropic-claude-ai-memory-system` - Claude AI memory system
- `anthropic-pdf`, `anthropic-xlsx`, `anthropic-docx`, `anthropic-pptx` - Document handling
- And more...

### OpenAI GPT (42 skills + 39 agents)
- `openai-gpt-5-thinking` - GPT-5 thinking mode
- `openai-gpt-5-1-professional` - Professional personality
- `openai-gpt-5-1-friendly` - Friendly personality
- `openai-gpt-5-1-cynical` - Cynical personality
- `openai-gpt-5-1-candid` - Candid personality
- `openai-gpt-4o` - GPT-4o system prompt
- `openai-gpt-4-1` - GPT-4.1 system prompt
- `openai-codex` - Codex system prompt
- `openai-chatgpt-gpt-5-agent-mode` - GPT-5 Agent Mode
- And 30+ more...

### Google Gemini (11 skills + 11 agents)
- `google-gemini-3-pro` - Gemini 3 Pro complete system prompt
- `google-gemini-3-flash` - Gemini 3 Flash
- `google-gemini-workspace` - Gemini Workspace integration
- `google-gemini-2-5-pro-guided-learning` - Guided learning mode
- `google-gemini-cli-system-prompt` - Gemini CLI
- `google-notebooklm-chat` - NotebookLM chat system
- And more...

### Perplexity AI (2 skills + 2 agents)
- `perplexity-voice-assistant` - Voice assistant system
- `perplexity-comet-browser-assistant` - Comet browser assistant

### xAI Grok (5 skills + 5 agents)
- `xai-grok-4` - Grok 4 system prompt
- `xai-grok-3` - Grok 3 system prompt
- `xai-grok-4-1-beta` - Grok 4.1 Beta
- `xai-grok-personas` - Grok persona system
- `xai-grok-com-post-new-safety-instructions` - Safety guidelines

### Proton (2 skills + 2 agents)
- `proton-luma` - Luma AI system
- `proton-lumo-2025-8-26` - Lumo 2025 version

### Misc AI Tools (8 skills + 8 agents)
- `misc-kagi-assistant` - Kagi search assistant
- `misc-notion-ai` - Notion AI system
- `misc-raycast-ai` - Raycast AI assistant
- `misc-le-chat` - Le Chat system
- `misc-confer` - Confer AI
- `misc-sesame-ai-maya` - Sesame AI Maya
- `misc-fellou-browser` - Fellou browser AI
- `misc-warp-2-0-agent` - Warp 2.0 agent

## How Ralph Auto-Triggers These

Ralph's agent selector automatically chooses the appropriate system prompt based on:

1. **Task Type Analysis**
   - Coding tasks → `anthropic-claude-code` or `openai-codex`
   - Document processing → `anthropic-pdf`, `anthropic-xlsx`, etc.
   - Voice/chat → `perplexity-voice-assistant`, `openai-gpt-4o-advanced-voice-mode`
   - Visual/image → `google-gemini-2-5-flash-image-preview`
   - Research → `openai-tool-deep-research`, `google-gemini-workspace`

2. **Personality Requirements**
   - Professional → `openai-gpt-5-1-professional`
   - Friendly → `openai-gpt-5-1-friendly`
   - Technical → `openai-gpt-5-1-nerdy`
   - Creative → `openai-gpt-5-1-quirky`

3. **Model Capabilities**
   - Thinking/reasoning → `anthropic-claude-4-1-opus-thinking`, `openai-gpt-5-thinking`
   - Fast execution → `google-gemini-3-flash`, `openai-o4-mini`
   - Deep reasoning → `anthropic-claude-3-7-sonnet`, `openai-gpt-5-2-thinking`

## Manual Usage

You can also manually invoke specific system prompts:

```
# Use Claude Code's actual system prompt
/skill anthropic-claude-code "Help me build a React app"

# Use GPT-5's thinking mode
/skill openai-gpt-5-thinking "Solve this complex problem"

# Use Gemini 3 Pro
/skill google-gemini-3-pro "Explain quantum computing"

# Use a specific personality
/skill openai-gpt-5-1-friendly "Write a casual email"
```

## Agent Mode

For full autonomous agent creation with these system prompts:

```
# Create an agent using Claude Code's system prompt
/agent anthropic-claude-code-agent

# Create an agent using GPT-5's system prompt
/agent openai-gpt-5-thinking-agent

# Create an agent using Gemini's system prompt
/agent google-gemini-3-pro-agent
```

## Ralph Integration Details

When Ralph analyzes a task, it:

1. **Scans available system prompts** from all categories
2. **Matches task requirements** to prompt capabilities
3. **Selects optimal system prompt** for the sub-task
4. **Applies prompt automatically** during execution

Example:
```
User: /ralph "Build a full-stack app with documentation"

Ralph's autonomous process:
1. Architecture phase → uses `anthropic-claude-3-7-sonnet` for complex reasoning
2. Backend coding → uses `anthropic-claude-code` for code generation
3. Documentation → uses `anthropic-docx` or `google-gemini-workspace`
4. Testing → uses `openai-codex` for test generation
5. Final review → uses `anthropic-claude-4-1-opus-thinking` for validation
```

## Adding More Prompts

To add new system prompts from the repository:

```bash
cd /tmp
git pull https://github.com/asgeirtj/system_prompts_leaks.git
python3 /tmp/convert_prompts.py
```

This will automatically:
- Detect new prompts
- Create skills and agents
- Make them available to Ralph

## Directory Structure

```
~/.claude/
├── skills/ralph-leaks/     # 91 skills (system prompts)
│   ├── anthropic-*.md
│   ├── openai-*.md
│   ├── google-*.md
│   ├── perplexity-*.md
│   ├── xai-*.md
│   ├── proton-*.md
│   └── misc-*.md
└── agents/ralph-leaks/     # 88 agents (full system prompts)
    ├── anthropic-*-agent/
    │   └── AGENT.md
    ├── openai-*-agent/
    │   └── AGENT.md
    └── ...
```

## Security Note

These are **actual leaked system prompts** from production AI systems. They contain:
- Behavioral guidelines
- Safety protocols
- Tool usage patterns
- Response formatting rules
- Personality configurations

Use them responsibly and in accordance with each provider's terms of service.

## Contributing

Found new system prompts? Submit them to:
https://github.com/asgeirtj/system_prompts_leaks

Then run the converter to add them to Ralph's toolkit.

---

**Total Integration: 179 system prompt skills and agents from major AI providers, auto-available to Ralph's autonomous agent selection.**