The Agent Skills Directory

COMMAND_EXECUTION (SAFE): The patterns.md file includes Python code snippets that utilize subprocess.check_output to capture Git metadata (HEAD hash and dirty status). These implementations use list-based arguments without shell=True, which is a secure way to execute specific system commands for metadata gathering.
EXTERNAL_DOWNLOADS (SAFE): The skill references standard package managers (pip, conda) and data versioning tools (dvc). It correctly demonstrates best practices for supply chain security by recommending pinned versions and content hashes for dependencies.
DATA_EXFILTRATION (SAFE): While the skill discusses data versioning with remote storage (S3), it uses these as illustrative examples for research data management. No patterns for exfiltrating sensitive system files or credentials were found.
DYNAMIC_EXECUTION (SAFE): The code patterns use yaml.safe_load() for configuration files, which is the recommended secure method for parsing YAML and prevents arbitrary code execution during deserialization.

data-reproducibility