The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from existing Google Slides presentations which could potentially contain malicious instructions.
Ingestion points: Commands such as read-slide, locate, and raw extract text and structure from slides. Additionally, insert-image and replace-slide allow the inclusion of content from external public URLs.
Boundary markers: The documentation proactively instructs the agent to use the --wrap-untrusted flag when reading content to help distinguish data from instructions.
Capability inventory: The skill has broad permissions to modify user data, including creating, deleting, and editing presentation elements (create, delete-slide, replace-text, element).
Sanitization: The tool includes a built-in mechanism via the --wrap-untrusted flag to encapsulate potentially dangerous content before the agent processes it.

gog-slides