The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/check_embedding_device.py and scripts/check_env.py to inspect system hardware and verify environment configuration. These calls target standard utilities such as sysctl, vm_stat, nvidia-smi, and git with hardcoded arguments, ensuring they are used only for local diagnostic purposes.- [REMOTE_CODE_EXECUTION]: Automated detectors flagged HTTP requests in scripts/check_embedding_device.py. Analysis shows these calls use urllib.request.urlopen solely to query local loopback addresses (localhost, 127.0.0.1) for Ollama and llama.cpp model catalogs. These interactions are restricted to local discovery and do not facilitate remote code execution from untrusted sources.- [DYNAMIC_EXECUTION]: scripts/scaffold_pipeline.py creates local support files by copying existing assets from the skill directory. The generated scripts rely on reputable Python packages such as sentence-transformers and do not perform unsafe operations like arbitrary code evaluation or runtime injection.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external files and APIs. It mitigates injection risks by defining ingestion points in specific playbooks, recommending the use of privacy filters and stable source identifiers, and requiring a consolidated review of all planned mutations before they are executed. This evidence chain provides sufficient controls for handling untrusted content during pipeline setup.

polygres-data-pipeline