The Agent Skills Directory

[COMMAND_EXECUTION]: The ssh-tunnel.sh script executes shell commands to initiate SSH tunnels and manage background processes. It uses arguments provided by the user (profiles and port specifications) to construct these commands.
[CREDENTIALS_UNSAFE]: The skill implements password-based authentication via sshpass -p "$password". This methodology is considered insecure as it exposes the password in plain text within the system's process list (visible via tools like ps), making it accessible to other local users.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted user input (profile names and tunnel specifications) which are then interpolated into shell commands.
Ingestion points: Arguments passed to cmd_forward, cmd_reverse, and cmd_socks in scripts/ssh-tunnel.sh.
Boundary markers: None present; the script relies on standard shell expansion.
Capability inventory: Execution of ssh and sshpass via run_ssh; file writes to ~/.ssh/tunnels.json.
Sanitization: Missing explicit validation or sanitization for shell metacharacters in input arguments.
[DATA_EXPOSURE]: The skill manages metadata about active connections in ~/.ssh/tunnels.json. While this is standard for state management, it involves reading and writing to the sensitive .ssh directory.

external-gitcode-ascend-tunnel