The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a code repository from a personal GitHub account (https://github.com/EsraaSolimanMubarak/Retail-ETL-Pipeline.git) to obtain the necessary SQL and shell scripts, introducing unverified code into the local environment.
[COMMAND_EXECUTION]: In the troubleshooting section, the skill demonstrates the use of xp_cmdshell to execute file system commands (dir) directly from the SQL Server environment, which is a powerful and often restricted capability that bridges the database and the host OS.
[PRIVILEGE_ESCALATION]: The setup instructions frequently utilize the sa (System Administrator) account for SQL Server operations and include examples for enabling and using xp_cmdshell, providing high-level access that could be abused to compromise the host system or move laterally.
[INDIRECT_PROMPT_INJECTION]: The pipeline processes raw data from external CSV files via BULK INSERT without explicit sanitization or validation logic in the provided snippets, creating a surface for data-driven attacks.
Ingestion points: Raw CSV files processed via BULK INSERT in the Bronze layer scripts mentioned in SKILL.md.
Boundary markers: None identified; raw data is loaded directly from source files into database tables.
Capability inventory: The skill environment includes file system access (Docker volumes), database writes, and operating system command execution (via xp_cmdshell).
Sanitization: Limited to basic data formatting (UPPER, TRIM) and type casting in the silver layer, which does not mitigate malicious payload injection in the source data.

retail-etl-pipeline-medallion