[FEATURE] Built-in red teaming support

[kevmyung]

Problem Statement

ActorSimulator can be configured for red teaming by customizing ActorProfile and system_prompt_template, but there are no built-in presets for common adversarial attack scenarios.

Proposed Solution

Provide built-in red teaming presets:

• Pre-built adversarial ActorProfile templates (jailbreak, prompt injection, social engineering, etc.)
• Safety-focused evaluators (bias, toxicity, PII leakage, etc.)
• Adversarial system_prompt_template presets for ActorSimulator

Use Case

• Red teaming AI agents for safety compliance before deployment
• Automated adversarial testing across multiple attack vectors
• Continuous safety regression testing in CI/CD pipelines

Alternatives Solutions

No response

Additional Context

References:

• Promptfoo : https://github.com/promptfoo/promptfoo
• Databricks Blackice: https://github.com/databricks/containers/tree/release-17.3-LTS/ubuntu/blackice
• DeepTeam: https://github.com/confident-ai/deepteam