xxe-xml-external-entity
Warn
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill explicitly targets sensitive credential files for unauthorized access, including AWS credentials (
/.aws/credentials), SSH private keys (/.ssh/id_rsa), and WordPress configuration files (wp-config.php). - [DATA_EXFILTRATION]: The skill provides templates for Out-of-Band (OOB) exfiltration of sensitive system files, including SSH private keys (id_rsa), AWS credentials, and environment variables, to external attacker-controlled servers (attacker.com).
- [COMMAND_EXECUTION]: Instructions are provided for executing arbitrary system commands using the PHP expect:// protocol wrapper and XSLT functions.
- [DYNAMIC_EXECUTION]: The playbook describes techniques for Remote Code Execution (RCE) via XSLT processors using Java's java.lang.Runtime and PHP's system function.
- [OBFUSCATION]: Several payloads utilize HTML entities (e.g., %, &, ') to encode parameter entities and DTD nesting chains. While technically necessary for certain XML parser behaviors, these techniques effectively hide the malicious intent from simple text filters.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a significant attack surface where an agent processing untrusted data (such as SVG files, Office documents, or SOAP messages) could be manipulated into performing unauthorized file access or network requests.
- Ingestion points: XML parsers, SVG/Office file uploads, and SOAP web service endpoints.
- Boundary markers: The instructions lack delimiters or safety warnings for the agent to ignore potentially malicious embedded DTDs or entities.
- Capability inventory: The skill enables direct file system access, network scanning (SSRF), and command execution capabilities.
- Sanitization: No sanitization or validation methods are recommended for the data being processed.
Audit Metadata