visa-doc-translate

Warn

Audited by Gen Agent Trust Hub on Mar 24, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes shell commands directly using sips for image processing and system package managers (pip, brew) for dependency installation.
  • [REMOTE_CODE_EXECUTION]: The skill implements a workflow where it dynamically generates a Python script containing data extracted from user-provided images via OCR and then executes that script to generate a PDF. This pattern of code generation from untrusted data can lead to arbitrary code execution if the extracted text contains malicious code fragments that are not properly escaped before interpolation.
  • [EXTERNAL_DOWNLOADS]: The skill downloads and installs several third-party libraries including easyocr, reportlab, pillow, and pytesseract from public repositories.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. 1. Ingestion points: Untrusted data enters the agent context via OCR text extraction from user-uploaded document images (e.g., bank statements, passports) in SKILL.md. 2. Boundary markers: The instructions do not specify any delimiters or warnings to ignore instructions embedded within the extracted text. 3. Capability inventory: The agent has the ability to execute shell commands (sips, pip, brew) and run dynamically generated Python code. 4. Sanitization: There is no evidence of sanitization or escaping of the extracted text before it is used in the translation or code generation steps.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 24, 2026, 07:42 AM
Security Audit — agent-trust-hub — visa-doc-translate