The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Python scripts scripts/auth.py and scripts/gmail.py to handle OAuth authentication and Gmail operations.
[DATA_EXFILTRATION]: The skill reads sensitive email content, headers, and labels, and provides commands like send and create-draft that could be used to exfiltrate this information if the agent is compromised or misled.
[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection. 1. Ingestion points: The skill reads full email bodies through scripts/gmail.py get. 2. Boundary markers: No specific boundary markers or instructions to ignore embedded commands are present in the skill description to protect the agent from malicious instructions within emails. 3. Capability inventory: The skill can send emails, create drafts, and modify labels, which allows an attacker to execute unintended actions via a malicious email. 4. Sanitization: No sanitization, filtering, or validation of email content is documented before it enters the agent's context.

gmail-automation