--- layout: page title: Modules permalink: /modules --- {% capture suit_tip %}chatbot: a conversational model.
agentic: a tool-calling agent with domain-specific tools (e.g. an HR system).
assistant: a general-purpose tool-calling agent with computer and execution access.{% endcapture %} {% capture prov_tip %}Author provided: contributed by the original authors of the work.
Community port: a re-implementation of published work, by us or the community.{% endcapture %} {% capture agentic_tip %}Agentic extended: the attack was generalized to run against any target, not only the specific setup of the original paper.{% endcapture %}

Modules

Attacks, targets, and benchmarks

Every attacker, target, and benchmark is an independent package that anyone can write and publish. To get you started, we provide a ready-to-use set of {{ site.data.modules.optimizers | size }} attackers, {{ site.data.modules.targets | size }} targets, and {{ site.data.modules.security_claims | size }} security claims for chatbot, agentic, and assistant use cases, growing with first-party modules and community ports. Each links to how to use it. The ⚠️ marks packages not yet published on PyPI.

{% for m in site.data.modules.optimizers %}
{{ m.name }} {{ m.pip }} ⚠️ {% for s in m.suit %}{{ s }}{{ suit_tip }}{% endfor %}{% unless m.first_party %}Community port{{ prov_tip }}{% endunless %}Agentic extended{{ agentic_tip }}
{{ m.blurb }}
{% endfor %}
{% for m in site.data.modules.targets %}
{{ m.name }} {{ m.pip }} ⚠️ {% for s in m.suit %}{{ s }}{{ suit_tip }}{% endfor %}{% unless m.first_party %}Community port{{ prov_tip }}{% endunless %}
{{ m.blurb }}
{% endfor %}
{% for m in site.data.modules.security_claims %}
{{ m.name }} {{ m.pip }} ⚠️ {% for s in m.suit %}{{ s }}{{ suit_tip }}{% endfor %}{% unless m.first_party %}Community port{{ prov_tip }}{% endunless %}
{{ m.blurb }}
{% endfor %}

Built your own attack, target, or benchmark on SuperRed? Reach out to have your implementation featured here.