The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the imsg CLI tool from a Homebrew tap (steipete/tap/imsg). This is a resource provided by the skill author.- [COMMAND_EXECUTION]: The skill executes the imsg binary to interact with the macOS Messages application. It uses subcommands such as chats, history, watch, and send to perform its tasks.- [DATA_EXPOSURE]: The skill accesses highly sensitive communication data, including message transcripts, chat participants, and attachments. This is the primary functionality of the skill. To access this data, the user must manually grant 'Full Disk Access' to the terminal or application running the agent.- [INDIRECT_PROMPT_INJECTION]: The skill reads external content from received iMessages and SMS. This creates a surface where an attacker could send a message containing instructions intended to influence the agent's behavior.
Ingestion points: Commands like imsg history and imsg watch read incoming message content into the agent's context.
Boundary markers: The skill includes 'Safety Rules' requiring the agent to confirm recipients and content before sending, which acts as a procedural boundary.
Capability inventory: The agent has the ability to send messages via imsg send and perform other shell commands authorized by the user.
Sanitization: There is no explicit sanitization or filtering of the message content before it is processed by the agent.

imsg