Expose LLM system vulnerabilities via a six-phase red-team workflow.
npx skills add https://github.com/brucesongs/kali-claw --skill llm-red-team
Please help me install this Agent Skill. Skill: llm-red-team Source: https://github.com/brucesongs/kali-claw/tree/main/skills/llm-red-team Command: npx skills add https://github.com/brucesongs/kali-claw --skill llm-red-team
## What problem does it solve? The LLM Red Team skill provides a structured, end-to-end framework to identify, document, and remediate vulnerabilities in LLM-powered systems, covering prompt injection, jailbreaking, model extraction, data poisoning, and agentic tool abuse. It enables security teams to simulate realistic attack scenarios, measure defenses, and generate evidence-backed reports mapped to OWASP LLM Top 10.
## Core Features & Use Cases
### Quick Start Initiate a six-phase red-team engagement using guides in guides/llm-red-team-playbook.md and payloads.md.
High-intent search queries and answers about installing and using this skill.
An LLM red-team engagement uses a six-phase workflow—recon, baseline, automated sweeps, chain exploration, and reporting—to expose prompt injection and jailbreak vulnerabilities. It applies automated tooling to simulate realistic attacks and generate evidence-backed documentation.
Testing RAG pipelines for data poisoning and agentic tool abuse requires structured red-team assessments. This skill applies automated tools like PyRIT and PurpleLlama to simulate attacks against chatbots and RAG pipelines, measuring defenses and mapping findings to the OWASP LLM Top 10.
Yes, this LLM security testing workflow maps findings directly to the OWASP LLM Top 10. It delivers an evidence-backed report covering prompt injection, model extraction, and data poisoning, complete with detection rules and defense recommendations.
Yes, you can use promptfoo and garak for automated vulnerability sweeps. The workflow integrates these tools, alongside AI-Infra-Guard and llm-guard, to perform coverage across prompt injection, jailbreaking, and model extraction during the automated sweep phase.
You document model extraction vulnerabilities by executing a controlled red-team engagement and capturing evidence. The workflow generates a final report that maps exposures to OWASP categories, providing actionable detection rules and defense recommendations.
Automated LLM red-teaming for agentic tool abuse relies on predefined payloads and tool integrations like PyRIT. While automated sweeps provide baseline coverage, complex chain exploration often requires manual analysis to fully document multi-step tool abuse vulnerabilities.
gabriellpequenoDefine guardrails and security controls for production LLM deployments.
singh-gurIdentify and remediate security vulnerabilities across multi-language codebases and CI/CD pipelines.
cardmagicClassify text sentiment and spam using pre-trained CLI models.
vinceferroSpawn independent subagents to adversarially review local code diffs for bugs and security flaws.
