linkfox-ruiguan-detection-patent-design
Warn
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The script
scripts/ruiguan_detection_patent_design.pydynamically modifies the Python module search path usingsys.path.insertto load shared modules from computed relative directories. - [PERSISTENCE]: The
references/onboarding.mdfile provides manual instructions for the user to append environment variable exports to persistent shell configuration files such as~/.zshrcand~/.bashrc. - [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external patent data including titles, abstracts, and specifications. 1. Ingestion points: Data is fetched via the search API in
scripts/ruiguan_detection_patent_design.py. 2. Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions embedded within the fetched patent content. 3. Capability inventory: The skill environment allows for local file writing and execution of Python scripts. 4. Sanitization: There is no evident sanitization of the retrieved content before it enters the model's context. - [COMMAND_EXECUTION]: The skill relies on the execution of provided Python scripts (
upload_image.py,onboarding.py,ruiguan_detection_patent_design.py) to perform its primary functions. - [EXTERNAL_DOWNLOADS]: The documentation suggests installing well-known third-party libraries (
requests,qrcode,pillow) viapipif they are missing from the environment.
Audit Metadata