linkfox-zhihuiya-pdf-data

Fail

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: HIGHPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PERSISTENCE]: The skill includes instructions in references/onboarding.md for the agent to modify user shell configuration files (~/.zshrc and ~/.bashrc) to store the LINKFOX_AGENT_API_KEY. Appending environment variables to shell profiles is a persistence mechanism that allows sensitive credentials to remain active across all future sessions.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of external data.
  • Ingestion points: scripts/zhihuiya_pdf_data.py processes patent IDs and publication numbers provided by the user and retrieves patent metadata and download links from an external API (tool-gateway.linkfox.com).
  • Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within the data retrieved from the Zhihuiya API.
  • Capability inventory: The skill possesses network capabilities (via urllib.request and requests) and file-writing capabilities (to the linkfox/ directory tree).
  • Sanitization: While the script uses standard JSON parsing, it does not implement sanitization or filtering of the API response before presenting it to the agent/user context.
  • [COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/onboarding.py and scripts/zhihuiya_pdf_data.py) which are responsible for managing account registration, billing flows, and API interactions. These scripts perform active network operations and manage local file systems.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 15, 2026, 07:11 AM