The Agent Skills Directory

[COMMAND_EXECUTION]: The send command in scripts/agent-msg uses tmux send-keys followed by an 'Enter' signal. This enables one agent session to execute arbitrary shell commands in any other active tmux session without session-level authorization.
[DATA_EXFILTRATION]: The read command in scripts/agent-msg uses tmux capture-pane to retrieve the terminal output/history of other sessions. This creates a risk for harvesting sensitive information like API keys, credentials, or environment variables that may have been displayed or logged in those sessions.
[PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface. Ingestion points: The read command in scripts/agent-msg ingests untrusted terminal content. Boundary markers: No markers are used to delimit external content or warn the agent. Capability inventory: The skill can inject commands (send), read terminal buffers (read), and access filesystem paths (info). Sanitization: Terminal output is retrieved and returned to the agent without any sanitization or filtering.

agent-comms