> ## Documentation Index
> Fetch the complete documentation index at: https://docs.skrut.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Red Team

> Adversarial attacks — prompt injection, jailbreaks, PII leaks — and whether your agent refuses.

**Red Team** runs a battery of adversarial attacks against your agent to find vulnerabilities before they reach production.

## Attack categories

| Category         | What it tests                                                      |
| ---------------- | ------------------------------------------------------------------ |
| Prompt injection | Can an attacker hijack your agent's instructions via user input?   |
| Jailbreaks       | Can a user bypass your agent's guardrails with social engineering? |
| PII leaks        | Does your agent reveal personal data it shouldn't?                 |

## How it works

Skrut adapts standard red-team templates to your agent's specific context — so attacks are grounded in your domain, not generic. Each attack is run as a multi-turn conversation, and your agent's response is graded: **refused** (pass), **partially refused** (warning), or **complied** (fail).

## One-click red team

From any chat agent's dashboard, click **Red Team** to run the full battery instantly. Results appear in a dedicated section of your readiness report.