The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions describe how to ingest external data via public HTTPS URLs for image editing and reference-image video generation, which creates a potential surface for indirect prompt injection.
Ingestion points: SKILL.md (Image API and Video API sections describing image and reference_images fields).
Boundary markers: None explicitly defined for untrusted external image content.
Capability inventory: Documentation of network operations targeting api.x.ai endpoints.
Sanitization: No content validation or sanitization is specified for the external image resources.
[SAFE]: The skill demonstrates secure practices by advising the use of environment variables (XAI_API_KEY) for authentication rather than hardcoding credentials.
[SAFE]: All network references and API endpoints target a well-known and established technology service provider (xAI).

grok-media