AI Agent System Prompts Library
What makes agentic coding applications different? TBH, it's mostly their system prompts, the tools they provide, and how they actually use them. In this repo, I'm documenting these system prompts and tools from different agentic coding solutions.
Table of Contents
Agent Comparison
| Agent | Provider | Type | Tools |
|---|---|---|---|
| Claude Code | Anthropic | CLI | 15 |
| Gemini CLI | CLI | 11 | |
| Cline | Open Source | VS Code | 12 |
| Aider | Open Source | CLI | Git-based |
| Roo Code | RooCode Inc | VS Code | 12 |
| Zed | Zed Industries | Editor | 15 |
| Codex CLI (Rust) | OpenAI | CLI | 1 |
What's Included
Each agent directory contains:
- System Prompts - Complete prompt templates with role definitions and instructions
- Tool Documentation - Detailed API specifications for all available tools
- Source Attribution - Direct links to original sources with retrieval dates
- README - Overview with GitHub repository links and feature summaries
Research Value
This collection provides insights into:
- Prompt Engineering Patterns - How production systems structure AI instructions
- Tool Design - Common patterns in AI agent tooling and capabilities
- Safety Measures - Security considerations and defensive programming practices
- User Experience - How different agents handle interaction and workflow management
Contributing
- Create a new directory under
agents/[agent-name]/ - Add a
system-prompt.mdfile with the complete system prompt - Document tools in
agents/[agent-name]/tools/directory - Include source information and retrieval date in headers
- Follow the established template formats (see CLAUDE.md)
Roadmap
Planned additions to this repository:
- Tools Comparison - Detailed analysis and comparison of tools across different agents
- Prompts Analysis - In-depth analysis of prompt engineering patterns and techniques
License
This collection is provided for research and educational purposes. Original prompts and tools remain property of their respective creators. See individual agent directories for specific source attributions.