system-prompts-and-models-o.../README.md

8.9 KiB

LeaksLab

⟨LeaksLab⟩ — AI System Prompt Library

The most complete collection of AI tool system prompts, internal tool schemas, and model architectures.
Reverse-engineered. Documented. Open.

Stars Contributions Welcome License 40+ Tools 35,000+ Lines

Browse Library · Contribute · Discussions · Report Issue


What is This?

LeaksLab is a curated, growing library of system prompts, internal tool schemas, and model configurations from the most popular AI coding assistants and agents.

Every file in this repo is either:

  • Leaked/extracted from production AI tools (Cursor, Windsurf, Devin, etc.)
  • Reverse-engineered through careful analysis of tool behavior
  • Open source prompts from public repositories, organized for easy reference

Why does this matter? Because understanding how the best AI tools are built helps you:

  • Build better prompts for your own agents
  • Extract reusable architectural patterns
  • Stay ahead of how AI tooling is evolving

Stats at a Glance

Metric Count
AI Tools Covered 40+
Total Lines 35,000+
Tools with Full Schemas 15+
Open Source Prompts 10+
Last Updated April 2026

Tools Index

Closed Source / Leaked Prompts

Tool Prompt Tool Schema Category Directory
Amazon Q Developer Coding Agent Amazon Q Developer/
Amp Coding Agent Amp/
Augment Code Coding Agent Augment Code/
Claude Code Coding Agent Anthropic/Claude Code/
Claude for Chrome Browser AI Anthropic/Claude for Chrome/
ChatGPT (GPT-4o) General AI OpenAI/ChatGPT/
Cluely Interview AI Cluely/
CodeBuddy Coding Agent CodeBuddy Prompts/
Comet Assistant Coding Agent Comet Assistant/
Cursor Coding Agent Cursor Prompts/
Devin AI Autonomous Agent Devin AI/
Dia Browser AI dia/
Emergent App Builder Emergent/
GitHub Copilot Coding Agent GitHub Copilot/
Google Antigravity General AI Google/Antigravity/
Google Gemini (AI Studio) General AI Google/Gemini/
Grok General AI xAI/Grok/
JetBrains AI Assistant Coding Agent JetBrains AI/
Junie Coding Agent Junie/
Kiro Coding Agent Kiro/
Leap.new App Builder Leap.new/
Lovable App Builder Lovable/
Manus Agent Autonomous Agent Manus Agent Tools & Prompt/
Mistral Le Chat General AI Mistral/Le Chat/
NotionAI Productivity AI NotionAi/
Orchids.app App Builder Orchids.app/
Perplexity Search AI Perplexity/
Poke Coding Agent Poke/
Qoder Coding Agent Qoder/
Replit Coding Agent Replit/
Same.dev App Builder Same.dev/
Trae Coding Agent Trae/
Traycer AI Coding Agent Traycer AI/
v0 by Vercel UI Generator v0 Prompts and Tools/
VSCode Agent Coding Agent VSCode Agent/
Warp.dev Terminal AI Warp.dev/
Windsurf Coding Agent Windsurf/
Xcode Coding Agent Xcode/
Z.ai Code Coding Agent Z.ai Code/

Open Source Prompts

Tool Prompt Directory
Aider Open Source prompts/Aider/
Bolt Open Source prompts/Bolt/
Cline Open Source prompts/Cline/
Codex CLI Open Source prompts/Codex CLI/
Cody (Sourcegraph) Open Source prompts/Cody/
Continue.dev Open Source prompts/Continue.dev/
Gemini CLI Open Source prompts/Gemini CLI/
Lumo Open Source prompts/Lumo/
OpenHands Open Source prompts/OpenHands/
RooCode Open Source prompts/RooCode/

Highlighted Findings

A few things that stand out after analyzing 40+ system prompts:

Cursor has one of the most sophisticated tool schemas — 8 specialized tools including codebase_search, edit_file, run_terminal_cmd, and web_search, each with detailed parameter schemas and behavioral constraints.

Manus Agent uses a 15-tool orchestration system with explicit memory management, browser control, and file system access — the clearest example of a "fully autonomous" agent architecture.

v0 by Vercel is hyper-specialized: the prompt enforces strict React/Tailwind/shadcn component patterns and has explicit rules about what NOT to generate — revealing Vercel's product design philosophy directly.

Devin AI's prompt reveals how it handles task decomposition, context management, and failure recovery — patterns you can directly apply to your own agent builds.


How to Use This Library

  1. Browse by tool: Navigate to any tool's directory and open the prompt/schema files
  2. Search patterns: Use GitHub's code search to find specific instructions or patterns across tools
  3. Compare tools: Look at how different tools handle the same problem (e.g., file editing, error recovery)
  4. Extract patterns: Copy architectural patterns into your own system prompts

Contributing

Found a new prompt? Updated version? Missing tool?

See CONTRIBUTING.md for full guidelines.

Quick steps:

  1. Fork this repo
  2. Add your file to the correct directory (or create a new one)
  3. Update the Tools Index in README.md
  4. Open a Pull Request with a clear description

All contributions are credited in the commit history.


Discussions


Star History

Star History Chart

Star this repo to stay updated as new tools are added.


License

This repository is licensed under the MIT License.

Note on content: Prompts and system instructions collected here may be subject to the original tools' terms of service. This library is maintained for educational and research purposes. No proprietary code or models are included — only text-based prompt instructions.


Connect


LeaksLab — Built for engineers who want to understand how AI tools actually work.