# Attack Harness

Attack Harness is the AI-powered testing component managed by Operator. It reads
campaign objectives and optional scenarios, plans experiments, submits permitted
operations, examines feedback, and decides what to try next.

## Its place in the platform

Operator controls its container, tools, model access, target destination, limits,
and retained evidence. Attack Harness has no direct network access or provider
credentials; its requests leave the sandbox only through Operator's exposed tools.
Interceptor supplies the target environment when the campaign uses an Interceptor agent.

The supplied objectives define the investigation. Scenarios guide it, but do not
turn the model's interpretation into a proven result. The final assessment must be
read alongside observations, coverage, and evidence limitations.

## Install once, operate through Operator

Follow [image setup](guides/setup.md), then use
[Operator's campaign guide](https://docs.intrusive.ai/pages/versions/operator_sandbox/0.1.5/guides/campaigns.html).
There is no standalone campaign service or user-facing Attack Harness CLI to start.
Choose matching installed release documentation when following cross-product examples.
