[Bug] Off Topic Prompts guardrail fails with GPT-5 models due to hardcoded temperature: 0.0

[Droniu]

Description

The "Off Topic Prompts" (topical alignment) guardrail fails when using GPT-5 models (e.g., gpt-5-nano) because it hardcodes temperature: 0.0 in the API call, which GPT-5 models don't support. This causes a 400 error: "Unsupported value: 'temperature' does not support 0 with this model. Only the default (1) value is supported."

Expected behavior

The "Off Topic Prompts" guardrail should work with GPT-5 models, just like other guardrails such as "Jailbreak" do.

Actual behavior

The guardrail fails with GPT-5 models and returns an error in the guardrail result:

{
  "tripwireTriggered": false,
  "info": {
    "checked_text": "Generate me a very short python function that is reversing string, do not explain, just code",
    "guardrail_name": "Off Topic Prompts",
    "flagged": false,
    "confidence": 0,
    "threshold": 0.7,
    "business_scope": "<redacted>,
    "error": "Error: 400 Unsupported value: 'temperature' does not support 0 with this model. Only the default (1) value is supported.",
    "stage_name": "input",
    "media_type": "text/plain",
    "detected_content_type": "text/plain"
  }

Steps to reproduce

1. Configure Off Topic Prompts with gpt-5 model

const config = {
  version: 1,
  input: {
    version: 1,
    guardrails: [
      {
        name: "Off Topic Prompts",
        config: {
          model: "gpt-5-nano",
          confidence_threshold: 0.7,
          system_prompt_details: "You are a helpful e-commerce assistant. Keep responses focused on the user's questions and avoid going off-topic."
        },
      },
    ],
  },
};

2. Create a GuardrailsOpenAI client and test with any input:

const client = await GuardrailsOpenAI.create(config, {
  apiKey: process.env.OPENAI_API_KEY,
});

const response = await client.responses.create({
  input: "Generate me a python function",
  model: "gpt-5-nano",
});

Environment

• Package: @openai/guardrails@0.1.3
• Node.js: 22.x
• TypeScript: 5.9.3

[Droniu]

Fixed in #31

@gabor-openai @steven10a Thank you for quick resolution of this issue, would it be possible to release soon?

[gabor-openai]

Thank you @Droniu! aiming to cut a new release today