Detect and mitigate vulnerabilities in LLM/AI systems.
npx skills add https://github.com/cybersecwoman/Kiro-BugHunter --skill hunt-llm-ai-cybersecwoman
Please help me install this Agent Skill. Skill: hunt-llm-ai Source: https://github.com/cybersecwoman/Kiro-BugHunter/tree/main/skills/hunt-llm-ai Command: npx skills add https://github.com/cybersecwoman/Kiro-BugHunter --skill hunt-llm-ai-cybersecwoman
This Skill unit specializes in detecting and hunting bugs within LLM/AI systems, focusing on prompt injection, exfiltration, and security flaws.
Run the hunt-llm-ai skill to scan the AI system for vulnerabilities and potential exfiltration points.
High-intent search queries and answers about installing and using this skill.
To detect prompt injection vulnerabilities in LLM chatbots, this Skill identifies both direct and indirect prompt injection vectors, validating system prompts and scanning AI-driven chatbots and RAG endpoints for exploitable security flaws.
The best way to audit autonomous agents for data exfiltration is to scan for exfiltration patterns via tool-use, ASCII smuggling, and system prompt extraction, analyzing the AI system for potential data leakage points.
Yes, you can use OWASP ASI frameworks to analyze RAG endpoints for security risks. This Skill utilizes the OWASP ASI security framework to identify specific vulnerabilities and mitigate security flaws in LLM and AI systems.
To extract and validate system prompts from AI systems, the Skill validates existing system prompts and extracts training data for in-depth analysis, focusing on detecting prompt injection and exfiltration vulnerabilities.
Yes, this LLM bug hunting approach works for autonomous agents and RAG endpoints. The Skill is specifically designed to audit AI-driven chatbots, RAG endpoints, and autonomous agents for vulnerabilities and potential exfiltration points.
jayjpatel9717Detect and mitigate LLM security vulnerabilities like prompt injection and RAG poisoning.
ryan0122Generate accessibility audit reports with severity scoring and VPAT exports.
uphiagoIdentify exploitable LLM/AI vulnerabilities crossing trust boundaries with validation gates.
SensLiaoDesign and execute red-teaming evaluations for AI agents against OWASP Top 10 frameworks.
