The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Gmail emails and Grain transcripts to generate draft replies, creating a potential surface for indirect prompt injection attacks.
Ingestion points: External data is ingested via the gog tool (Gmail messages) and the mcporter tool (Grain meeting transcripts) as defined in the drafting logic in SKILL.md.
Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions to isolate the external content from the agent's core logic.
Capability inventory: The skill possesses the ability to write to the file system via the Write tool and execute shell commands via Bash for logging, Gmail interaction, and WhatsApp notifications.
Sanitization: There is no evidence of sanitization or filtering applied to the ingested content before it is processed by the model.
[COMMAND_EXECUTION]: The skill frequently executes local scripts and specialized tools including python3, gog, and mcporter to perform logging, auditing, and platform interactions.
[DATA_EXFILTRATION]: The skill transmits potentially sensitive email draft content and metadata to a configured WhatsApp number. While this is the primary functionality intended for user review, it represents the transfer of communication data to an external messaging service.

email-drafting