The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the user conversation, creating a surface for potential indirect prompt injection attacks.
Ingestion points: Untrusted data enters the agent context through the detected_files (filenames) and known_urls parameters, which are sourced from the active conversation.
Boundary markers: There are no explicit delimiters or boundary markers used when interpolating these values into the tool call instructions.
Capability inventory: The skill interacts with the picsart_upload_widget and mp_upload tools and subsequently feeds data into downstream media processing tools.
Sanitization: No specific sanitization or validation of the input strings is performed by the skill instructions.
[SAFE]: The skill implements a security-conscious architecture for handling local files. By delegating the upload process to a browser-based widget, the skill ensures that the agent never has direct access to the user's local filesystem, mitigating risks related to data exposure and unauthorized file access. The external resources (picsart.io) and tool names are consistent with the identified vendor, Picsart.

picsart-add-media