linkfox-zhihuiya-bibliography
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves patent titles, abstracts, and applicant names from the external Zhihuiya database. This data is untrusted and provides a surface for indirect prompt injection where malicious instructions could be embedded in patent metadata to influence the agent. \n
- Ingestion points: Patent bibliography API responses processed in
scripts/zhihuiya_bibliography.py.\n - Capability inventory: File writes to local directories and network operations for API requests.\n
- Boundary markers: Instructions focus on structured output but lack explicit warnings to ignore instructions within the retrieved content.\n
- Sanitization: No specific sanitization or filtering of patent text is implemented before presentation to the user.\n- [PERSISTENCE]: The documentation in
references/onboarding.mdinstructs the agent to provide users with shell commands to modify their profiles (e.g.,~/.zshrc,~/.bashrc) to store API keys. This facilitates persistence of credentials across shell sessions.\n- [EXTERNAL_DOWNLOADS]: Thescripts/onboarding.pyscript checks for dependencies such asrequestsandqrcodeand instructs the user to install them from the Python Package Index (PyPI) if they are not already available.\n- [COMMAND_EXECUTION]: The skill requires the execution of local Python scripts to communicate with vendor APIs and manage local cache files. These scripts handle session identifiers and authentication tokens.
Audit Metadata