Identify exploitable LLM/AI vulnerabilities crossing trust boundaries with validation gates.
npx skills add https://github.com/uphiago/recon-skills --skill hunt-llm-ai
Please help me install this Agent Skill. Skill: hunt-llm-ai Source: https://github.com/uphiago/recon-skills/tree/main/redteam/hunt-llm-ai Command: npx skills add https://github.com/uphiago/recon-skills --skill hunt-llm-ai
This Skill solves the common pain point of wasting time on confabulated false positives when testing LLM/AI features, providing a structured validation framework to identify only exploitable vulnerabilities that cross trust boundaries and meet professional reporting standards.
Use the hunt-llm-ai skill to test the target's LLM-powered chatbot and related AI features for exploitable prompt injection and exfiltration vulnerabilities, validating all findings with out-of-band callbacks to avoid false positives.
High-intent search queries and answers about installing and using this skill.
To eliminate false positives during LLM prompt injection pentesting, validate findings using a 5-point gate that requires out-of-band callbacks, run-twice reproducibility, and verifiable cross-tenant artifacts. This prevents reporting model confabulation as actual vulnerabilities.
Testing RAG systems for data poisoning involves identifying exploitable vulnerabilities that cross trust boundaries aligned with OWASP LLM standards. You validate potential poisoning flaws by verifying cross-tenant artifacts and anchored leaks to ensure the issue is genuinely exploitable.
Yes, you can find cross-tenant IDOR vulnerabilities in agentic copilots by testing AI data layers for trust boundary crossings. The validation framework requires verifiable cross-tenant artifacts to confirm the IDOR is exploitable and meets professional reporting standards.
To validate system prompt leakage during an AI security assessment, use anchored leaks and run-twice reproducibility. This confirms the leaked data is a genuine system prompt extraction rather than a confabulated response from the target LLM.
Yes, this method covers OWASP Agentic Applications ASI01-ASI10 security flaws. It tests agentic copilots and MCP tools for exploitable vulnerabilities, applying a structured validation framework to eliminate false positives and ensure findings meet professional reporting standards.
Out-of-band callbacks are needed for markdown exfiltration testing to prove data exfiltration crosses trust boundaries. This validation technique confirms the LLM actually transmitted data externally, distinguishing true exploitable flaws from non-exploitable model behavior or confabulation.
sparkfabrikAudit AI agent configurations and LLM integrations for OWASP Top 10 security risks.
pdparchitectDetect prompt injection and agentic tool misuse against OWASP ASI standards.
dreadnodeGuide data exfiltration payload crafting for AI rendering pipelines.
ricardorocheDetect prompt injection attempts in AI and LLM applications.
