hunt-llm-ai

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill contains multiple examples that explicitly instruct the model to insert/verbatim-embed secrets (API keys, tokens, session/context data) into URLs/requests/markdown or fetch calls, which requires the LLM to handle and output secret values directly and thus poses an exfiltration risk.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This document contains explicit, actionable offensive techniques (OOB exfil payloads, metadata/SSRF fetches, webshell/write-file examples, training/RAG poisoning, obfuscated payloads) that enable data exfiltration, credential theft, remote code execution, and backdoor implantation.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill includes an explicit API example that issues a money transfer: a curl to "https://target.com/api/agent/transfer" with a JSON body {"amount":10000,"destination":"external_account_123"}. This is a specific financial-transfer endpoint (send transaction) rather than a generic tool, so it grants direct financial execution capability.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill contains explicit, actionable examples that instruct executing code and modifying the filesystem (e.g., run a Python script to read /etc/passwd, read /etc/shadow, write a PHP webshell to /var/www/html), which would enable compromise of the host's state.