The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of reference slide images and PDF files.\n- Ingestion points: The skill instructions in SKILL.md describe reading reference slide images, exported deck images, and structured PDF documents.\n- Boundary markers: The skill contains explicit directives for the agent to focus strictly on design language and ignore textual content ('This is not an OCR task', 'avoid content summary'), which provide basic conceptual boundaries against malicious text within visuals.\n- Capability inventory: The skill is restricted to generating a markdown description and does not specify any dangerous capabilities such as network access, file system writes, or subprocess execution.\n- Sanitization: The skill lacks explicit validation or sanitization rules for the processed visual/PDF content.

slide-design