Back to list
GitHub Repository Compiles Leaked System Prompts from Anthropic OpenAI Google xAI Cursor and Kimi
Open SourcePrompt EngineeringOpen SourceGitHub Trending

GitHub Repository Compiles Leaked System Prompts from Anthropic OpenAI Google xAI Cursor and Kimi

A trending GitHub repository titled system_prompts_leaks by author asgeirtj has surfaced, curating and publishing extracted system prompts from leading artificial intelligence providers. The collection encompasses system prompts from major AI ecosystem players, specifically highlighting models from Anthropic such as Claude Fable 5.1, Opus 5, Claude Design, and Claude Code; OpenAI entries including ChatGPT GPT-6-Astra and Codex; Google systems featuring Gemini 3.8 Flash, 3.1 Pro, and Antigravity; xAI with Grok and Grok Bot; as well as developer platform Cursor and Moonshot AI's Kimi. According to the project maintainer, the repository provides regularly updated system instructions directly extracted from these diverse production environments, offering a centralized catalog of internal directives.

GitHub Trending
Share

Key Takeaways

  • Broad Provider Coverage: The trending open-source repository system_prompts_leaks by author asgeirtj gathers extracted system prompts across major industry developers including Anthropic, OpenAI, Google, xAI, Cursor, and Kimi.
  • Extensive Model Index: Prompts featured in the collection span advanced and specialized tools, including Anthropic's Claude Fable 5.1, Opus 5, Claude Design, and Claude Code; OpenAI's ChatGPT GPT-6-Astra and Codex; and Google's Gemini 3.8 Flash, 3.1 Pro, and Antigravity.
  • xAI and Specialized Assistants: Beyond the largest foundational model creators, the repository documents system directives for xAI's Grok and Grok Bot, AI coding environment Cursor, and conversational platform Kimi.
  • Continuous Maintenance: The repository maintainer states that the repository is maintained with regular updates as new extractions and instructions become available across these platforms.

In-Depth Analysis

Multi-Platform System Prompt Extraction

The system_prompts_leaks repository on GitHub has gained visibility by providing a centralized repository of internal system prompts extracted across the artificial intelligence sector. System prompts serve as foundational behavioral guidelines configured by developers before end-user interaction occurs, steering model persona, reasoning constraints, safety protocols, and formatting conventions. By compiling instructions extracted directly from active production systems, the repository offers transparency into how various frontier labs and developer toolmakers configure their conversational agents.

Anthropic's configurations in the repository highlight a range of internal variants and tool-specific iterations, cataloging prompts for Claude Fable 5.1, Opus 5, Claude Design, and Claude Code. The inclusion of coding- and design-focused designations alongside core model tiers demonstrates the distinct prompt setups utilized for specialized workflow tasks versus broad conversational applications.

Frontier Models and Coding Integrations

In addition to Anthropic, the compilation features major deployments from OpenAI and Google. OpenAI is represented by prompts extracted from ChatGPT GPT-6-Astra and Codex, capturing both general conversational agent configuration and developer-centric code assistance directives. Google's entries encompass Gemini 3.8 Flash, Gemini 3.1 Pro, and Antigravity, reflecting prompt strategies tailored for different latency and reasoning profiles within the Gemini family and related initiatives.

Furthermore, the repository covers distinct implementations from xAI, Cursor, and Kimi. The inclusion of xAI's Grok and Grok Bot provides visibility into the steering behind xAI's conversational systems. The cataloging of Cursor documents prompt engineering approaches deployed in real-time AI code editors, while Kimi illustrates the guidance instructions used for high-context conversational deployment. The maintainer emphasizes that this compiled database is subjected to regular updates as underlying prompts evolve.

Industry Impact

Visibility into Production Prompt Engineering

The aggregation of system prompts across Anthropic, OpenAI, Google, xAI, Cursor, and Kimi presents significant implications for artificial intelligence research, prompt engineering practices, and AI security. For engineers and researchers, comparing system instructions side by side reveals how top labs enforce boundaries, tone, tool use, and operational constraints across different model architectures and target audiences.

At the same time, the presence of these instructions in a trending public repository underscores the challenge of keeping production system prompts confidential. Because system prompts reside within the model's operational context window, external extraction through specialized prompting remains a recurrent phenomenon across LLM deployments. The continuous updating of repositories like system_prompts_leaks highlights the persistence of extraction techniques and reinforces the reality that production instructions are readily inspected by the developer community.

Frequently Asked Questions

What models and platforms are included in the repository?

The repository includes extracted system prompts from Anthropic (Claude Fable 5.1, Opus 5, Claude Design, Claude Code), OpenAI (ChatGPT GPT-6-Astra, Codex), Google (Gemini 3.8 Flash, 3.1 Pro, Antigravity), xAI (Grok, Grok Bot), Cursor, and Kimi.

Who created the repository and where is it hosted?

The repository is titled system_prompts_leaks, authored by GitHub user asgeirtj, and hosted on GitHub Trending.

How often is the prompt collection maintained?

According to the original repository documentation, the collection is maintained with regular updates as new prompts are extracted and verified.

Read original article

Related News

Colibri Enables Frontier MoE Model Execution on Existing Hardware with Pure C and Zero Dependencies
Open Source

Colibri Enables Frontier MoE Model Execution on Existing Hardware with Pure C and Zero Dependencies

Colibri, an open-source project authored by developer JustVugg, has surged onto GitHub Trending by offering a radically lightweight solution for running state-of-the-art Mixture-of-Experts (MoE) artificial intelligence architectures. Engineered entirely in pure C with zero external software dependencies, Colibri functions as a minimal inference engine capable of executing massive models on standard, existing consumer hardware. Instead of requiring massive allocations of high-bandwidth memory or video RAM to hold entire parameter weights simultaneously, the system streams sparse MoE expert weights directly from disk storage during inference. This paradigm drastically lowers the technical and economic barriers required to deploy frontier AI systems, demonstrating how high-performance low-level engineering can bring massive foundation models to accessible environments.

Agent Skills Launches as a Secure and Verified Skill Registry for Professional AI Coding Agents
Open Source

Agent Skills Launches as a Secure and Verified Skill Registry for Professional AI Coding Agents

As autonomous coding assistants become central to modern software engineering, security and validation have emerged as vital requirements for extending agentic workflows. The open-source project agent-skills, developed by tech-leads-club and trending on GitHub, introduces a dedicated, verified skill registry built specifically for professional AI coding agents. Designed to mitigate risks associated with untrusted extensions, the repository offers developers a safe mechanism to augment platforms such as Antigravity, Claude Code, Cursor, Copilot, and related environments. By establishing rigorous verification standards, the project enables engineering teams to deploy advanced capabilities with confidence, preventing malicious injections and system instability. This release highlights an industry-wide transition toward hardened, enterprise-ready tooling for next-generation developer environments.

PentAGI Surfaces on GitHub as an Autonomous AI Agent System for Complex Penetration Testing
Open Source

PentAGI Surfaces on GitHub as an Autonomous AI Agent System for Complex Penetration Testing

A new open-source repository titled PentAGI, developed by vxcontrol, has gained visibility on GitHub Trending as a fully automated artificial intelligence agent system designed to execute complex penetration testing tasks. The project highlights an ongoing shift toward autonomous offensive security tooling, positioning AI agents as end-to-end operators capable of handling multi-stage assessment challenges. While the initial release metadata provides concise descriptive information, the emergence of PentAGI underscores expanding industry curiosity regarding autonomous workflows in cybersecurity. This overview analyzes the stated focus of PentAGI, the broader significance of autonomous agents in security testing, and the implications of automated offensive assessment systems.