The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of googlemail-client, google-oauth, and caffeineai-authorization from the Mops package registry. While external automated scans flagged the documentation links for these packages as potential phishing risks, they are standard resources within the Motoko developer ecosystem.
[DATA_EXFILTRATION]: Facilitates the transmission of user-composed email content and OAuth authentication tokens to external Google Cloud endpoints.
[COMMAND_EXECUTION]: Instructs users to perform terminal-based package management using the mops CLI tool.
[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by accepting and processing untrusted text for email delivery without explicit sanitization.
Ingestion points: The sendEmail function in src/backend/mixins/gmail-messaging.mo accepts to, subject, and body parameters from the agent's context.
Boundary markers: No specific delimiters or "ignore instructions" warnings are implemented for the email body content.
Capability inventory: The skill utilizes the gmail_users_messages_send capability in src/backend/lib/gmail.mo to send messages to external recipients.
Sanitization: User-provided text is encoded to UTF-8 for the Gmail API but is not filtered for embedded instructions.
[SAFE]: The skill follows advanced security practices including the use of PKCE to secure the OAuth handshake, strict role-based access control for administrative configuration, and the use of non-replicated IC outcalls to prevent secret leakage across blockchain nodes.

connector-googlemail