nof1.ai-predictive-prompt

English | 中文

According to the API responses from https://nof1.ai/, use ChatGPT to infer the system_prompt based on the returned user_prompt, cot_trace, and llm_response.

Predictive System Prompt

system_prompt

Inference Prompt Retrieval Process

1. Request: https://nof1.ai/api/conversations to obtain the conversation list, for example:

{
  "conversations": [
    {
      "id": "claude-sonnet-4-5_1761529280.571673",
      "user_prompt": "xxx",
      "llm_response": {
        "ETH": {
          "quantity": 2.45,
          "stop_loss": 3945,
          "signal": "hold",
          "profit_target": 4445,
          "invalidation_condition": "4H MACD turns negative or price closes below $3,950 on 4H timeframe",
          "justification": "",
          "confidence": 0.65,
          "leverage": 10,
          "risk_usd": 350,
          "coin": "ETH"
        }
      },
      "cycle_id": 3075,
      "inserted_at": 1761529283.999278,
      "cot_trace": "xxx",
      "model_id": "claude-sonnet-4-5",
      "skill": "swing_trading",
      "cot_trace_summary": "xxx",
      "timestamp": 1761529280.570891,
      "run_id": "20251022131211"
    }
  ]
}

2. Retrieve user_prompt and cot_trace
3. Examine llm_response to confirm that the expected llm output is a JSON containing coins and their parameters
4. Since the API does not return system_prompt, infer the user_prompt. I used ChatGPT, DeepSeek, and Qwen respectively, and found ChatGPT's inference to be the most comprehensive.
5. Test: Send the system_prompt and user_prompt to DeepSeek-R1 or other models to check whether the returned result matches the one from the API.