The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process information from external URLs and uploaded files (PDFs, images, documents). This behavior constitutes a surface for indirect prompt injection, as the agent may encounter instructions embedded in the source material. However, this is the primary intended function of the skill.
Ingestion points: Content retrieved via web_fetch and data from user-uploaded brand guidelines in SKILL.md.
Boundary markers: The skill does not currently use explicit delimiters or specific instructions to the agent to ignore directives found within the external brand data.
Capability inventory: The skill uses web_fetch, web search, and writes files to the /mnt/user-data/outputs/ directory.
Sanitization: There is no explicit sanitization or instruction-filtering defined for the ingested brand content.
[SAFE]: All file operations and network requests for brand research (e.g., Google Fonts, university brand guides) are legitimate and consistent with the skill's stated purpose. The installation and build instructions in the documentation follow standard development practices.

brand-yml