The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted email content (messages, threads, attachments) retrieved via the Gmail API. Since the agent also possesses write capabilities (sending emails, trashing threads, modifying labels), this creates a surface where malicious instructions embedded in an incoming email could potentially influence the agent's behavior.
Ingestion points: SKILL.md (via curl requests to /messages/{id} and /threads/{id}).
Boundary markers: None present in the provided recipes to differentiate email content from instructions.
Capability inventory: Shell command execution (Bash), Gmail API read/write/send operations, and package installation (npm).
Sanitization: No specific filtering or sanitization of message bodies is implemented before the content is processed by the AI.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the Google Workspace CLI tool (@googleworkspace/cli) from the official NPM registry to facilitate advanced email operations.
[COMMAND_EXECUTION]: The skill utilizes shell commands (curl, jq, base64) and an external CLI to perform API requests and data processing.

google-gmail