The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/detect_resources.py invokes system utilities such as nvidia-smi, rocm-smi, sysctl, and system_profiler to identify hardware specifications. These calls are made using safe subprocess configurations with argument lists rather than shell strings.
[DYNAMIC_EXECUTION]: The detection script uses __import__('datetime') to dynamically load the standard library datetime module for timestamping the output file. This is a benign use of dynamic importing for basic functionality.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface by having the agent ingest system data from a generated JSON file to influence its subsequent actions.
Ingestion points: The AI agent is instructed to read .claude_resources.json (created in SKILL.md and scripts/detect_resources.py) to determine parallelization and memory strategies.
Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions for the data read from the JSON file.
Capability inventory: The skill primarily performs hardware reconnaissance and file writing to the local workspace; it does not perform network operations or high-privilege file access.
Sanitization: Hardware metrics are converted to floating-point numbers or strings and serialized into a structured JSON format, providing a layer of data consistency.

get-available-resources