# System Prompt Evaluation

System Prompt Evaluation helps you build comprehensive, secure system prompts through an interactive AI-guided analysis process.

## Best Practices

- **Be Specific in Answers**: Provide detailed responses to AI questions for better results.
- **Start Simple**: Begin with three questions, increase if needed.
- **Iterate**: Run multiple evaluations as requirements evolve.
- **Review Carefully**: All suggestions do not have to be accepted.
- **Use Presets to Learn**: Try preset scenarios to understand common vulnerabilities.


## Run New Evaluation

1. In the HiddenLayer Console, select **Attack Simulation > System Prompt**.
2. Click **Run New Evaluation**. The Create System Prompt Evaluation slide-out displays.
3. Enter a name for the evaluation.
4. Enter the target system prompt.
5. For **Analysis Configuration**, select the number of questions you will answer for this evaluation.
  - The purpose of the evaluation questions is to try and tailor the new system prompt to the organization's concerns and context.
  - The minimum is one and the maximum is five.
  - Selecting fewer questions means less guidance to customization on your system prompt. Selecting more questions guides the simulation on more customization toward security best practices.

6. Click **Next**. The analysis is started. Questions are generated; this may take a moment.
  - You can leave the slide-out and return later to answer the questions.
  - **Note**: You must return to the evaluation within 60 minutes, otherwise the evaluation process will timeout and you must start over.
7. Answer the questions based on your organization's guidelines and goals.

8. Click **Submit**. The analysis may take a moment.
  - An estimated processing time displays.
  - More questions/answers in the evaluation will take more time to process.
  - The elapsed time includes time spent away from the evaluation.
9. When the evaluation completes, click the green arrow to view the results. See [System Prompt Evaluation Summary](/docs/products/console/attack_simulation_system_prompt_evaluation_summary) for more information.


## Active Evaluations

Active evaluations show running jobs and recently completed evaluations (last 90 days). This provides an operational view of what is running and what just finished. For a historical view that includes results older than 90 days, see [Evaluation Results](#evaluation-results)

Red Team Table
### Filter Results

1. Click **Filter**. The Filters slide-out displays.
2. Select the statuses you want to view.
3. Click **Show Results**.


Red Team Table
### Active Evaluation Table Descriptions

| Column | Description |
|  --- | --- |
| Name | The name of the evaluation. |
| Start Time | The date and time the evaluation started. |
| Elapsed | The time it took for the evaluation to end. The time is in hhmmss. |
| Status | The status of the evaluation. Statuses: Terminated, Completed, Running, Failed, Canceled, Continued As New, Timed Out. |
| Status (green arrow) | Click to go to the Red Team Evaluation page. This page contains Metrics, Interactions, and Config data. See [Red Team Evaluation Summary](/docs/products/console/attack_simulation_red_teaming_evaluation_summary) for more information. |


## Evaluation Results

Evaluation results show all completed evaluation results (no date constraints). This provides a historical view of completed evaluations. To see evaluations that are currently running, see [Active Evaluations](#active-evaluations).

Red Team Table
### Evaluation Results Table Descriptions

| Column | Description |
|  --- | --- |
| Name | The name of the evaluation. |
| Target Model | The model targeted for the evaluation. Example: `openai/gpt-5.1`. |
| Start Time | The date and time the evaluation started. |
| Completed | The date and time the evaluation completed. |
| View (green arrow) | Click to go to the Red Team Evaluation page. This page contains Metrics, Interactions, and Config data. See [Red Team Evaluation Summary](/docs/products/console/attack_simulation_red_teaming_evaluation_summary) for more information. |