The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides instructions and Python code that access sensitive local credential paths.
Evidence: The Python code in SKILL.md explicitly reads from ~/.vast_api_key and references ~/.ssh/vastai_key for instance access.
[COMMAND_EXECUTION]: The skill includes code snippets that use the subprocess module to execute system commands.
Evidence: The secure_cleanup function in SKILL.md executes ssh and vastai binaries via subprocess.run.
[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface where untrusted data could influence shell commands, as it lacks sanitization for external inputs used in command construction.
Ingestion points: Function arguments instance_id, ssh_host, and ssh_port in the secure_cleanup function within SKILL.md.
Boundary markers: None present; the code does not include instructions to ignore embedded commands in these parameters.
Capability inventory: Uses subprocess.run to call ssh (which executes remote commands) and the vastai CLI tool.
Sanitization: Absent. Arguments are interpolated directly into shell strings (e.g., f"root@{ssh_host}") without escaping or validation beyond basic type casting.
[DYNAMIC_EXECUTION]: The skill demonstrates the dynamic assembly of shell command strings at runtime.
Evidence: The secure_cleanup function in SKILL.md uses f-strings to concatenate host information and remote commands before passing them to the system shell via ssh.

vastai-security-basics