The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted technical descriptions from the external Zhihuiya patent database, which could contain instructions intended to influence the agent.
Ingestion points: Technical patent text is fetched via the /zhihuiya/descriptionDataTranslated API and handled by scripts/zhihuiya_description_translated.py.
Boundary markers: The instructions do not define clear delimiters or "ignore" directives to isolate the retrieved patent content from the agent's core instructions.
Capability inventory: The skill uses Python scripts with the ability to write JSON data to the local workspace and perform network operations.
Sanitization: No explicit filtering or sanitization of the technical technical content is performed before it is rendered to the agent.
[PERSISTENCE_MECHANISMS]: The onboarding guide (references/onboarding.md) includes instructions for persisting authentication credentials across sessions.
Evidence: The documentation provides commands for users to append API keys to shell profiles such as ~/.zshrc and ~/.bashrc, or use setx on Windows.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The onboarding script references external Python packages that are required for specific features.
Evidence: scripts/onboarding.py suggests the installation of qrcode, pillow, and requests via the standard pip package manager from the official registry.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages authentication tokens and transmits technical identifiers to remote servers.
Evidence: The skill reads LINKFOX_AGENT_API_KEY from environment variables and communicates with tool-gateway.linkfox.com and api.linkfox.com. These endpoints are confirmed vendor-owned resources.

linkfox-zhihuiya-description-data-translated