The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes content from potentially untrusted PDF files via the display_pdf and get_text tools. If a document contains instructions disguised as content, it could influence the agent's subsequent actions.
Ingestion points: The display_pdf tool loads documents from external HTTPS URLs provided at runtime.
Capability inventory: The skill possesses the ability to extract document text (get_text), capture visual snapshots (get_screenshot), and apply modifications (add_annotations, fill_form).
Boundary markers: There are no explicit instructions for the agent to treat document content as untrusted data or use specific delimiters.
Sanitization: The skill description does not specify methods for sanitizing text extracted from the PDFs before it is processed by the agent.
External Resource Loading: The skill is designed to fetch documents (display_pdf) and image assets for annotations (add_annotations with type: "image") from remote HTTPS URLs. This interaction with external servers is a standard feature for document collaboration but represents a point of interaction with external data sources.

view-pdf