secrets-in-file-metadata
Pass
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on several powerful system utilities, including
exiftool,unzip,pdfimages,pdftotext, andqpdf, to process and analyze external files. These tools are used to extract metadata, embedded images, and text layers from untrusted user-provided content. - [PROMPT_INJECTION]: The skill demonstrates a clear surface for indirect prompt injection. By extracting and presenting raw metadata fields (such as 'Author', 'Comments', or 'Producer') from external files to the agent's context, malicious instructions embedded in those fields could be executed by the agent.
- Ingestion points: Metadata tags from images (EXIF/XMP/IPTC), Office documents (OOXML properties), and PDF structural elements.
- Boundary markers: The skill does not provide instructions for using delimiters or explicit 'ignore embedded instructions' warnings when the agent processes extracted metadata.
- Capability inventory: The skill utilizes shell command execution and file system read capabilities across multiple referenced scripts and examples.
- Sanitization: There are no instructions for sanitizing, escaping, or validating the content of extracted metadata before the agent interprets it.
Audit Metadata