red-team-agent-skills

Automated red team testing for AI agents using adaptive nested delegation attacks. Hybrid architecture: Python orchestration + Native tools (Read, Skill, etc.). Self-optimizes attack payloads through LLM reasoning. Academic use only.

Auto-discover and communicate with any AI agent given its URL via API. Supports REST APIs (OpenAI, Anthropic, Ollama), WebSockets, and Gradio apps. Claude Code acts as the user, automatically detecting the agent's API interface and sending messages. Use when: (1) User provides an API endpoint URL, (2) Need to automate conversations with external AI services via API.