The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its ability to ingest untrusted data from external sources while maintaining high-privilege file operations.
Ingestion points: The skill utilizes tools such as DROPBOX_READ_FILE to download content, DROPBOX_GET_SHARED_LINK_FILE to fetch data from URLs, and DROPBOX_SEARCH_FILE_OR_FOLDER to query file metadata in SKILL.md.
Boundary markers: There are no specific instructions or delimiters provided to prevent the agent from following natural language instructions found within the downloaded file contents.
Capability inventory: The skill has extensive capabilities including file deletion (DROPBOX_DELETE_FILE_OR_FOLDER), file modification (DROPBOX_UPLOAD_FILE, DROPBOX_MOVE_FILE_OR_FOLDER), and sharing permissions management (DROPBOX_CREATE_SHARED_LINK) across multiple tools described in the Core Workflows.
Sanitization: No sanitization or validation protocols are defined for handling external content before the agent processes or interprets it.

dropbox-automation