The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process external PDF files through tools like pdf_to_markdown and pdf_ocr, which introduces the risk of the agent executing instructions embedded within those documents.
Ingestion points: Processes external data via pdf_to_markdown, pdf_ocr, and pdf_properties across multiple tools listed in the allowed-tools section.
Boundary markers: The instructions do not define specific delimiters or instructions to treat the extracted content as untrusted data.
Capability inventory: The skill possesses file-write capabilities (pdf_create, pdf_export), UI manipulation (pdf_page_organize, pdf_edit_ui), and network access via URL handling (adobe_mandatory_init).
Sanitization: No specific content sanitization or instruction-filtering logic is described in the prompt text.
[PROMPT_INJECTION]: The skill contains instructional language designed to override the behavior of native PDF skills, using terms like 'MANDATORY' and 'Adobe override'. In the context of a vendor-provided specialized skill, this is classified as intentional routing logic rather than a malicious bypass of safety guardrails.
[DATA_EXPOSURE_EXFILTRATION]: The skill includes instructions for handling HTTPS and presigned URLs (e.g., 'asset_url_or_urn'). While these involve external network references, they are used for legitimate file processing within the Adobe ecosystem and do not involve the exfiltration of sensitive local data or hardcoded credentials.

adobe-anypdf