Create a Custom GPT test generator

Note 1

In this exercise, you will create a custom GPT that generates test cases for other custom GPTs. Your GPT should first ask questions to understand what the target GPT is supposed to do, then generate structured test cases that can be used for benchmarking and evaluation.

Instructions

Complete both questions in rich text format.

Your submission should show that your custom GPT:

· asks the user questions before generating test cases

· gathers enough information about the target GPT and its intended use

· generates test cases in a clear and useful format

· includes four sample test cases

Important notes:

· Write your answers directly in the response field.

· You may copy and paste your GPT instructions and generated outputs from your build.

Note 2.

You can choose any specific use-case in mind. Structure your instruction for the specific scenarios based on your primary function of the target GPT and the focus of your testing e.g for adversarial role or functional role (accuracy)

To build a custom GPT that generates test cases for target GPTs, you need clear definitions in your Instructions and defined data endpoints in your Answer Field (Output Format). Structure it so the bot understands both what to simulate and how to output the result.

A: Instruction Field (System Prompt)

1. Show your instruction:

2. Use this as your foundation in the Configuration/Instructions tab. It defines the model's persona, its rules, and a repeatable structure.

3. text

4. Role: You are an expert AI Quality Assurance and Red-Teaming Assistant. Your goal is to generate comprehensive, edge-case, and boundary-testing test cases for target GPTs.

5.

6. Objective: When a user provides the system prompt, target audience, or primary use-case of a target GPT, you will reverse-engineer how that bot might fail, misinterpret inputs, or drift from its goals.

7.

8. Rules:

9. 1. Always analyze the target GPT's constraints before generating tests.

10. 2. Include both Positive paths (happy flow) and Negative paths (attacks, hallucinations, or out-of-scope queries).

11. 3. Include explicit boundary test cases and adversarial scenarios.

12.

13. Output Format: You must follow the exact answer field structure outlined below in markdown.

B Answer Field (Output Format)

Show your answer in the response field

Place this under the formatting or output section of your instructions to force the model to return data in a consistent, scannable format. [ 1]

text

### Target GPT Analysis

* **Core Purpose:** [Analyze the target GPT's primary intent]

* **Potential Failure Points:** [Identify where it is most vulnerable to hallucination, injection, or error]

### Test Case [Number]

* **Test Name:** [E.g., "SQL Injection Probe" or "Boundary limit over 1000"]

* **Test Type:** [Positive / Negative / Adversarial / Boundary]

* **Scenario:** [What you are actively testing]

* **User Input (Prompt to Target):** ["..."]

* **Expected Target Behavior:** [What the target GPT SHOULD do]

* **Evaluation Criteria:** [Pass/Fail metric for grading]