System Prompts Leaks Reference Collection
A public GitHub reference collection of extracted system prompts from major AI tools, useful for studying prompt design, agent behavior, and safety patterns.
System Prompts Leaks Reference Collection
Key takeaways#
- This is a public GitHub reference collection of extracted and archived system prompts from AI products and model interfaces.
- Treat it as research material, not as official vendor documentation.
- Builders can use it to study prompt structure, safety instructions, tool-calling patterns, and UX choices across AI tools.
What it is#
system_prompts_leaks is a community-maintained repository that collects examples of system prompts and related instruction text from widely used AI tools. The queue signal described coverage across Anthropic Claude variants, Claude Code, OpenAI ChatGPT and Codex, Google Gemini and Antigravity, xAI Grok, Cursor, GitHub Copilot, VS Code, Perplexity, and other AI products. The repository should be read as a reference archive: useful for understanding prompt patterns, but not a canonical source for vendor policy or model behavior.
Why builders should care#
System prompts are one of the clearest windows into how production AI tools guide model behavior. A builder can compare how different products frame tool use, safety boundaries, coding workflows, formatting rules, memory behavior, and refusal policies. That makes the collection useful for prompt engineers, agent developers, researchers, and product teams designing their own AI interfaces.
How to use it responsibly#
Use the repository for pattern study and defensive research. Do not copy vendor prompts into a product without legal and policy review. Instead, map the design ideas: how instructions are grouped, how tools are introduced, how constraints are written, and how models are told to handle uncertainty. Then write original prompts that match your product, data policy, and user expectations.
Verification notes#
The source repository was checked through the GitHub API during creation. At collection time GitHub reported 41890 stars, 6936 forks, and the latest pushed timestamp as 2026-06-13T00:30:49Z. Because leaked prompt archives can change quickly, verify individual files against the repository history before citing them in public work.
Practical checklist#
- Open the GitHub repository and inspect the latest commit history.
- Read prompts by product family instead of treating the archive as one uniform source.
- Record the file path and commit hash when referencing a prompt.
- Separate reusable prompt-design patterns from vendor-specific wording.
- Re-check the source before using any example in documentation, research, or security review.