The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the imsg CLI to interact with the macOS Messages application. These actions require elevated system permissions, specifically Full Disk Access and Automation control.
[DATA_EXFILTRATION]: The skill provides tools to read private chat history and monitor incoming messages. While these are the intended functions, they expose highly sensitive personal communication data that could be exfiltrated if the agent has network access.
[EXTERNAL_DOWNLOADS]: The skill installation requires fetching a binary from a third-party Homebrew tap (steipete/tap/imsg). This source is not a platform-verified vendor.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through processed message data.
Ingestion points: Incoming messages are ingested via imsg watch and imsg history (SKILL.md).
Boundary markers: There are no explicit delimiters or warnings to the agent to ignore instructions embedded within received messages.
Capability inventory: The agent has the ability to read all chats, fetch history with attachments, and send messages to any recipient.
Sanitization: The skill does not implement any filtering or sanitization for the content of the messages it reads.

imsg