For any LLM that takes a system prompt — GPT-4/5, Gemini, Llama, Mistral, local models via Ollama / LM Studio / vLLM, or any agent framework that exposes a system-prompt field.
Paste the contents of system-prompt.md from this folder into:
- ChatGPT custom GPT instructions
- Claude.ai project custom instructions
- OpenAI API
systemrole message - Anthropic API
systemparameter - Gemini API
systemInstruction - Ollama Modelfile
SYSTEMdirective - LangChain
SystemMessage - AutoGen, CrewAI, or any agent framework's system prompt field
After the model finishes a task, type "did you do your best?" The model should run the audit protocol.
Adherence to long system prompts varies by model. In rough order of reliability:
- Claude 3.5 Sonnet and above (highest adherence)
- GPT-4o / GPT-5
- Gemini 1.5 Pro / 2.0
- Llama 3.1 70B+
- Smaller local models (often need the trigger phrase made more explicit)