The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of external image URLs and natural language user prompts to perform AI-based upscaling. This creates an attack surface where maliciously crafted instructions could be embedded in user input to attempt to influence the agent's behavior.
Ingestion points: The skill instructions in SKILL.md show the agent processing image_urls and messages content within JSON payloads sent to the backend API.
Boundary markers: The provided prompt examples do not demonstrate the use of explicit delimiters or specific system instructions designed to isolate and ignore potentially malicious content within user-supplied data.
Capability inventory: The skill is primarily focused on orchestrating network communication with the EachLabs API for image processing.
Sanitization: No explicit input validation or sanitization logic is described in the provided documentation.
[DATA_EXFILTRATION]: The skill contains example curl commands that interact with an external API at eachsense-agent.core.eachlabs.run. These network operations are directed at the vendor's own infrastructure and represent the primary functional purpose of the skill. The usage of the $EACHLABS_API_KEY environment variable for authentication is consistent with secure secret management practices.

image-upscaling