The Agent Skills Directory

[DATA_EXFILTRATION]: Communicates with an external, non-whitelisted backend domain (https://google-workspace-extension.geminicli.com) to handle OAuth token authentication and refresh flows.
[INDIRECT_PROMPT_INJECTION]: Processes untrusted data from external sources (Google Drive files, names, and folders) that could contain malicious instructions designed to manipulate the agent.
Ingestion points: File metadata and content queries fetched via scripts/drive.py through search and list operations.
Boundary markers: Absent; data returned from Google Drive is forwarded to the agent context without structural delimiters or instruction-isolation markers.
Capability inventory: File system write operations via download, as well as Google Drive write actions such as upload, create-folder, move, rename, and trash in scripts/drive.py.
Sanitization: Employs escape_query_string to protect against Google Drive native query injection, but lacks filtering or validation to neutralize prompt injection content intended for the LLM.

google-drive