The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes user-provided content to generate slides, including a subset of HTML and CSS. However, it contains explicit defensive instructions to mitigate these risks.
Ingestion points: User-provided content for slide text, cards, and tree-based or HTML/CSS layouts within SKILL.md.
Boundary markers: The skill explicitly defines a 'Codex capability boundary' in SKILL.md, instructing the agent to resolve tools against the host and not broaden access or invent tools.
Capability inventory: The skill uses tools for creating presentations, adding slides, composing layouts, inserting images, and exporting files (google_drive_export_file).
Sanitization: The skill specifies that for HTML/CSS layouts, 'Only the documented tags and properties compile; anything else is refused by name, never dropped,' which serves as a validation mechanism.
[SAFE]: The skill includes explicit security guidance, such as 'Never widen sharing or make a private image public to satisfy an image URL requirement,' which encourages the agent to maintain the user's existing security posture.
[SAFE]: References to external repositories in MIRROR.md point to well-known technology organizations (GitHub/cursor) for source tracking purposes and do not involve remote code execution or automated downloads.

cursor-google-slides