CL4R1T4S is a transparency project that collects extracted system prompts from major AI models and agents. Built by elder-plinius (Pliny).
"In order to trust the output, one must understand the input."
The project exposes: - What AIs can't say - What personas/functions they're forced to follow - How they're told to lie, refuse, or redirect - What ethical/political frames are baked in by default
Includes prompts from: OpenAI, Google, Anthropic, xAI, Perplexity, Cursor, Windsurf, Devin, Manus, Replit, and more.
The Claude Design prompt (ANTHROPIC/Claude-Design-Sys-Prompt.txt) is particularly valuable for understanding how to structure an AI design agent. Key patterns extracted → see claude-design-workflow-patterns.
Understanding how major platforms structure their agent instructions provides reusable patterns for building your own agent workflows. The system prompts reveal: - Workflow architecture patterns - How to structure variation generation - Context-first design principles - Progressive complexity approaches