The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file defines an ssh_exec Python pattern that uses subprocess.run to invoke the system ssh client. It explicitly uses the -o StrictHostKeyChecking=no flag, which disables the security feature that verifies the identity of the remote host, making the connection vulnerable to Man-in-the-Middle (MitM) attacks.
[INDIRECT_PROMPT_INJECTION]: The provided ssh_exec function pattern creates a vulnerability surface for command injection. If an agent or user adopts this code and passes untrusted strings to the cmd parameter, it would allow execution of arbitrary commands on the remote GPU instance.
Ingestion points: The cmd parameter in the ssh_exec function definition (SKILL.md).
Boundary markers: Absent. The pattern does not suggest the use of delimiters or provide warnings regarding the execution of user-supplied commands.
Capability inventory: The skill documents the use of subprocess.run for shell command execution.
Sanitization: Absent. The code pattern lacks input validation, shell escaping, or sanitization logic.

vastai-sdk-patterns