A GitHub repository containing the extracted system prompts of more than 28 major AI coding tools has reached 134,000 stars, providing developers with unprecedented insight into how these platforms actually operate. The collection, curated by developer Lucas Valbuena, exposes the internal instructions that tools like Cursor, Windsurf, Claude Code, Augment Code, and Devin AI send to their underlying language models.

System prompts function as the foundational layer that shapes AI tool behavior, defining priorities, refusal patterns, output formatting, and available tool calls. Before this repository emerged, developers had limited visibility into these critical components, relying primarily on marketing descriptions or extensive hands-on testing to understand tool capabilities.

The repository demonstrates remarkable depth beyond simple prompt collection. It includes JSON tool schemas that reveal the specific capabilities each vendor has chosen to implement, providing insight into strategic priorities and technical approaches. With 489 commits across 28 contributors and regular updates through March 2026, the project maintains relevance as tools evolve rapidly.

Examining the collected prompts reveals distinct competitive strategies. Windsurf's extensive collection spanning 11 "waves" of tool definitions suggests aggressive iteration cycles, likely driven by real-world failure modes encountered in production environments. Cursor's transition to "Agent Prompt 2.0" indicates substantial architectural changes rather than incremental improvements, pointing to fundamental user experience challenges that required comprehensive solutions.

Augment Code's inclusion of GPT-5 tool definitions provides confirmation of multi-model provider support that isn't prominently featured in their public documentation. This flexibility likely enables more sophisticated task routing based on model availability, cost considerations, or specific capability requirements.

Devin AI's DeepWiki prompt structure emphasizes knowledge retrieval capabilities beyond basic code generation, aligning with the autonomous agent positioning that requires reasoning across large codebases rather than completing isolated tasks.

The repository serves multiple constituencies effectively. For developers evaluating coding assistants, it provides technical details that marketing materials cannot match. Teams building AI development tools gain access to proven architectural patterns for context management, tool integration, and agent behavior that have already scaled to millions of users.

The version history functionality adds particular value by enabling analysis of how vendors adjust their approaches over time. These changes often reveal where companies encounter challenges, iterate most heavily, or quietly roll back unsuccessful experiments.

Security implications represent an increasingly important consideration. The repository maintainer explicitly addresses prompt extraction risks and references services designed to identify these vulnerabilities. What seemed theoretical a year ago now appears as a practical concern for AI startups that have treated their prompts as protected intellectual property.

For practical application, developers building VS Code extensions or similar tools can reference existing implementations rather than starting from scratch. Comparing Cursor's agent prompts with VSCode Agent approaches provides concrete examples of different architectural decisions and their trade-offs.

The broader competitive landscape reveals Claude's growing role as infrastructure rather than just a standalone product. Multiple tools in the collection use Claude as their backend inference layer, reflecting a trend toward specialization where companies focus on user experience and workflow integration while leveraging established model providers.

This transparency fundamentally alters how the AI coding tool market operates. Developers can make more informed decisions based on actual implementation details, while tool builders gain access to battle-tested patterns that can accelerate their own development cycles. The repository represents a shift toward greater openness in a space that has traditionally relied on proprietary advantages.

Related Links:

• Context Engine
• Pricing
• Docs
• Blog
• Sign in
• Book demo