read-metadata

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions utilize shell command templates containing placeholders for user-supplied paths and filenames (e.g., <file.jpg> and <target>). If the agent populates these placeholders with unsanitized input, it creates a potential command injection surface. However, this risk is significantly mitigated by the allowed-tools frontmatter which restricts shell execution to specific binaries like exiftool, find, and jq.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests metadata from external files and presents it to the agent for processing and summarization.
  • Ingestion points: Processes EXIF, IPTC, and XMP metadata from image files (SKILL.md).
  • Boundary markers: Absent; the skill does not use delimiters to isolate metadata values from instructional context.
  • Capability inventory: Includes access to exiftool, find, jq, Read, and Write (SKILL.md).
  • Sanitization: Absent; no escaping or validation is performed on the metadata content before it is displayed or parsed.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 06:24 PM
Security Audit — agent-trust-hub — read-metadata