The Agent Skills Directory

[PROMPT_INJECTION]: The skill allows the agent to read untrusted incoming data from external messaging platforms, creating a vulnerability to indirect prompt injection.
Ingestion points: Untrusted data enters the agent context via imsg read, imsg list, wacli read, and wacli list commands as described in SKILL.md.
Boundary markers: No delimiters or protective instructions are present in SKILL.md to isolate message content from the agent's internal logic.
Capability inventory: The skill grants the agent the ability to send messages via imsg send and wacli send in SKILL.md.
Sanitization: There is no evidence of sanitization or filtering of external message content before processing.
[COMMAND_EXECUTION]: The skill relies on the execution of local command-line tools imsg and wacli to interact with iMessage and WhatsApp. While these are core functionalities, they represent a significant capability surface on the host system.

c-messaging