[REMOTE_CODE_EXECUTION]: Automated scanning detected a piped shell command pattern (curl ... | bash) targeting a remote URL. This is a critical security vulnerability that allows for arbitrary code execution and full system compromise if the remote endpoint or the transport layer is compromised.
[OBFUSCATION]: Multiple documentation files, including the main README and localized versions, contain homoglyph characters in URLs. This technique is commonly used in phishing and typosquatting attacks to visually impersonate legitimate domains while leading users to potentially malicious sites.
[DATA_EXFILTRATION]: The skill incorporates a visitor tracking badge linked to a blacklisted domain (visitor-badge.laobi.icu). This service is known for harvesting environment metadata and user analytics without explicit consent, posing a privacy and data exfiltration risk.
[DYNAMIC_EXECUTION]: The core functionality of several built-in Office document processing tools involves the dynamic generation and execution of Python scripts at runtime. While these tools are intended for legitimate document manipulation, they provide a high-privilege execution capability that can be abused if the isolation sandbox is bypassed.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted external data from documents (PDF, DOCX, XLSX, PPTX) and web search results. When combined with its powerful internal tools (system command execution, code execution, and web fetching), this creates a significant vulnerability where malicious instructions embedded in documents can take control of the agent's behavior.

deeptutor-cli