The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external image URLs, creating an attack surface where instructions embedded within analyzed images could influence the agent's behavior.
Ingestion points: Image data retrieved via the imageUrl parameter and processed through the multimodal API in scripts/multimodal_recognize_image.py.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are included in the prompt logic.
Capability inventory: The skill can execute Python scripts, write to the linkfox/ directory, and perform network operations to vendor endpoints.
Sanitization: No explicit text sanitization or filtering is performed on the analysis results before they are returned to the agent.
[PERSISTENCE]: The onboarding documentation in references/onboarding.md provides manual instructions for users to persist their API credentials by adding export commands to shell profile files such as .bashrc, .zshrc, and .profile.
[COMMAND_EXECUTION]: The skill relies on the execution of local Python scripts (upload_image.py, multimodal_recognize_image.py, and onboarding.py) to handle image processing and account tasks.
[DATA_EXFILTRATION]: The scripts/upload_image.py script is designed to upload local files to the LinkFox OSS cloud to generate publicly accessible URLs for analysis, which is the declared primary function for handling local images.
[EXTERNAL_DOWNLOADS]: The skill interacts with official vendor domains (linkfox.com) for service delivery, authentication, and billing operations, which is consistent with the vendor's intended functionality.

linkfox-multimodal-recognize-image