Show HN: A little catalogue of system prompts (api.chat)

A GitHub “little catalogue of system prompts” collects dozens of provider-specific system prompts and instruction packs from major LLM vendors and third parties—Anthropic (Claude series and UserStyle modes), OpenAI (ChatGPT/4o/4.1/5-, GPT-4.5, Codex variants), Google DeepMind (Gemini), Mistral, xAI (Grok family), plus Bolt, Replit, Manus, Cursor, Perplexity and many smaller projects. Entries include prompt lengths and a “View” link; sizes range from compact utilities (e.g., a 55-word GPT-4o Image Gen postfill) to very large system prompts (several over 8K–9.7K words such as Claude 4, Sonnet 3.7 New, Devin2). The catalogue covers diverse prompt types—coding assistants, tool/integration stacks, persona and behavior modes, functions/tool specs and image/voice helpers. For the AI/ML community this is both a practical resource and a signal: it’s a rich dataset for prompt-engineering research, transferability studies, instruction-tuning and benchmarking of model alignment and robustness. Researchers can analyze structure, token budgets, and common scaffolding used by production systems, while practitioners can reuse orchestration patterns for tool use, safety hooks, or multi-step reasoning. At the same time it raises IP, security and misuse questions—exposed proprietary prompt strategies, potential for adversarial replication, and licensing risks—so responsible handling, provenance checks, and red-teaming are essential when leveraging the collection.