The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess module to execute ssh, scp, and ssh-keygen commands. Specifically, scripts/ssh_exec.py allows executing arbitrary shell commands on remote hosts using the ssh command. While these are intended functions, they provide a powerful interface for command execution.
scripts/ssh_exec.py: Uses subprocess.run to execute commands on remote aliases.
scripts/ssh_transfer.py: Uses subprocess.run to execute scp for file transfers.
scripts/ssh_tunnel.py: Uses subprocess.call to establish SSH tunnels.
[DATA_EXFILTRATION]: The skill facilitates the transfer of files between the local machine and remote servers using scp via scripts/ssh_transfer.py. This capability can be used to move sensitive data if the agent is directed to do so without proper oversight.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (SSH configuration files and remote command outputs) and interpolates this data into the agent's context. There is a risk that malicious content in ~/.ssh/config or output from a compromised remote server could influence the agent's behavior.
Ingestion points: Reads ~/.ssh/config in scripts/ssh_config.py and captures stdout/stderr from remote commands in scripts/ssh_exec.py.
Boundary markers: The scripts return structured JSON, which helps delimit data, but the SKILL.md does not specify explicit boundary markers for the AI's prompt when handling these outputs.
Capability inventory: Includes file read/write (~/.ssh/config), network operations (SSH/SCP), and remote command execution.
Sanitization: The scripts use json.dumps for output, but there is no specific sanitization of the remote command output before it is presented to the agent.

ssh