The Agent Skills Directory

[COMMAND_EXECUTION]: Executes various shell commands to inspect system hardware and remote compute status.
Evidence: Uses nvidia-smi and ssh to query GPU memory usage and connectivity.
Evidence: Invokes vendor-specific CLI tools including vastai and modal to verify infrastructure availability.
Evidence: Executes dynamic Python snippets to verify PyTorch CUDA/MPS compatibility.
[PROMPT_INJECTION]: Vulnerable to indirect prompt injection through unvalidated project configuration files.
Ingestion points: Reads target environment settings and server hostnames from CLAUDE.md (Step 1).
Boundary markers: Absent. The skill does not implement delimiters or instructions to ignore malicious commands embedded within the config file.
Capability inventory: The skill has permissions to execute ssh and python based on the parsed configuration.
Sanitization: No explicit validation or escaping of hostnames or environment variables is performed before interpolation into shell command templates.

aris-compute-guard