The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides several scripts (exec.sh, connect.sh) that facilitate arbitrary remote command execution on hosts via SSH. This is a primary feature of the skill but inherently carries risk if the agent is directed to execute malicious strings.
[COMMAND_EXECUTION]: The scripts take user-provided arguments (hostname, command, ports) and interpolate them directly into shell commands. While used within a controlled administrative context, this presents a surface for command injection if input is not carefully handled by the calling agent.
[DATA_EXFILTRATION]: The skill allows the agent to read and potentially modify sensitive local SSH configuration files (~/.ssh/known_hosts, ~/.ssh/config) and internal memory files (~/clawd/MEMORY.md).
[SAFE]: The skill consistently uses -o StrictHostKeyChecking=no and -o UserKnownHostsFile=/dev/null. While this is a security best practice violation in standard environments as it enables Machine-in-the-Middle (MitM) attacks, it is documented here as a workaround for dynamic IP addresses within a private Tailscale network, which is a common (though risky) administrative pattern.
[DATA_EXFILTRATION]: The check-host.sh script performs port scanning on the target remote IP, which could be used for internal network reconnaissance.

ssh-manager