The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill establishes a mechanism for agents to send messages to one another, which inherently creates an indirect prompt injection surface.
Ingestion points: The target agent receives arbitrary text input via the letta CLI commands described in SKILL.md.
Boundary markers: The skill mentions a <system-reminder> tag which acts as a delimiter, but this is a soft boundary that does not strictly prevent instruction injection from the sender agent.
Capability inventory: Target agents are described as having the ability to execute their own tools, access memory, perform web searches, and make API calls.
Sanitization: There is no evidence of sanitization or escaping of the message content within the provided skill instructions.
[COMMAND_EXECUTION] (LOW): The skill relies on executing shell commands via the letta CLI.
The commands (letta agents list, letta messages search, and letta -p) are used for legitimate administrative and communication tasks within the Letta ecosystem.
[SAFE] (NO_CODE): The skill contains no executable scripts or binary files, consisting entirely of metadata and instructional markdown documentation.

messaging-agents