The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of @primitivedotdev/cli and @primitivedotdev/sdk. These are official packages associated with the skill's authoring vendor, used to provide the email functionality.
[COMMAND_EXECUTION]: The skill provides detailed instructions for executing shell commands such as primitive send and primitive sending. These commands interact with the Primitive relay and allow the agent to manage message delivery and scheduling. The skill also explicitly warns against insecure practices like passing secrets in command-line arguments.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its message auditing features.
Ingestion points: The primitive sent list and primitive sent get commands retrieve email records (including subject and body content) from the Primitive relay service, which may originate from untrusted external sources.
Boundary markers: The instructions do not define any specific delimiters or boundary markers to isolate retrieved email content from the agent's instructions.
Capability inventory: The agent has capabilities including sending emails, reading local files via the --body-file and --attachment flags, and scheduling future operations through the CLI.
Sanitization: There is no mention of sanitization, filtering, or validation processes to verify the content of retrieved emails before the agent processes them.

primitive-send