AI tool system prompts hit 131K GitHub stars: what developers should know

A GitHub repository collecting the full system prompts, internal tools, and model configurations of over 30 AI coding tools has reached 131,000 stars and 33,300 forks. The repo covers nearly every major player in the AI-assisted development space, including Cursor and Windsurf, as well as Claude Code, Augment Code, and Devin. For developers evaluating these tools or building their own, it is now the largest publicly available reference on how commercial AI agents actually work under the hood.

What Happened

The repository, maintained by user x1xhlol (Lucas Valbuena), has accumulated 486 commits across 28 contributors since its creation. According to the repo description, it contains over 30,000 lines of prompt content spanning system prompts, tool definitions, and model configurations for tools including Cursor, Windsurf, Claude Code, Augment Code, Devin AI, Replit, Lovable, Manus, v0, Kiro, Junie, VSCode Agent, Xcode, Warp.dev, and many others.

The collection is licensed under GPL-3.0. Commits show active maintenance: the most recent README update landed on March 9, 2026, with tool-specific prompt files updated across the preceding months. A few notable entries:

• The Augment Code directory includes a gpt-5-tools.json file added in August 2025.
• Cursor's directory contains an "Agent Prompt 2.0" file from November 2025.
• Windsurf's section includes tool definitions up through "Wave 11."

Key Features

• Full system prompts for 30+ tools. Complete extracted prompts for Cursor, Windsurf, Augment Code, Devin AI, Replit, v0, Lovable, Manus, Junie, Kiro, and more. Not summaries: the actual instruction text these agents receive.
• Tool and function definitions included. Several directories contain JSON files defining the internal tools available to each agent, showing exactly what capabilities each product exposes to its underlying model.
• Model configuration details. The repo reveals which models power specific features. The Amp directory, for instance, includes prompts for both Sonnet and GPT-5 configurations.
• Version history via Git. Because the repo tracks changes over time, you can diff prompt versions. Cursor's agent prompt reached version 2.0, and Windsurf's tool definitions evolved through at least 11 waves.
• Cross-platform coverage. The collection spans Cursor, Windsurf, VSCode Agent, Xcode, Devin, Manus, Lovable, Replit, v0, Same.dev, Perplexity, NotionAI, and more.
• Open source prompt section. A separate directory collects prompts from tools that publish theirs voluntarily.

Why It Matters

System prompts are the invisible layer that defines how an AI coding tool behaves: what it prioritizes, what guardrails it follows, and how it structures tool calls. Having these in the open lets developers make informed comparisons rather than relying on marketing copy.

For teams building internal AI tools or custom agents, this repo is a practical reference library. You can study how Devin structures multi-step task execution, how Cursor handles agent-mode tool selection, or how Windsurf defines its function schemas, then apply those patterns to your own systems.

It also raises real questions about prompt security. The repo itself links to a service called ZeroLeaks, noting that "exposed prompts or AI models can easily become a target." If your product's competitive advantage lies in a system prompt, this repository serves as a reminder that prompts are extractable.

Example Use Case

Say you're building a coding agent that wraps Claude Sonnet 4.6 for a Python monorepo. You want the agent to edit files, run tests, and explain changes, but you're unsure how to structure the system prompt and tool definitions.

You pull Cursor's Agent Prompt 2.0 and Augment Code's GPT-5 tools JSON from the repo. From Cursor, you adapt the instruction structure for handling multi-file edits and user confirmations. From Augment Code's tool definitions, you model your own JSON schema for shell execution and file read/write.

You cross-reference with the Anthropic directory to see how Anthropic instructs Claude to handle code generation. In an afternoon, you have a well-structured prompt informed by three production-grade implementations, rather than starting from scratch.

Competitive Context

The repo makes side-by-side comparison straightforward. According to the repo's commit history, Cursor and Windsurf, the two most popular AI code editors, take notably different approaches in their agent prompts: Cursor's directory shows a versioned agent prompt structure, while Windsurf organizes around numbered "waves" of tool definitions.

Claude Code and Augment Code differ in scope: Claude Code's prompts (in the Anthropic directory, updated as recently as March 2026 with a Sonnet 4.6 file) focus on Anthropic's own coding assistant, while Augment Code's directory includes model-specific tool configurations. Devin AI, positioned as a fully autonomous agent, has a separate DeepWiki prompt file that suggests a more complex architecture.

All six tools, including Cursor, Windsurf, Claude Code, Augment Code, and Devin, are represented with distinct prompt strategies, making this a useful resource for understanding where each product draws its boundaries.

Bottom Line

This repository gives developers a detailed look at how commercial AI coding tools instruct their models. If you're choosing between tools, building your own agent, or just curious about what sits between the LLM and your IDE, it's worth a read. At 131K stars and growing, the developer community has clearly decided this kind of transparency has value.