The Agent Skills Directory

Shell Command Execution via Notebooks: The skill provides a technical workaround that involves executing a shell command (!echo) within a Spark notebook environment. This command is designed to append configuration settings to a hidden file (.trident-context) located in a service user's home directory (/home/trusted-service-user/). While described as a method to enable specific catalog features in legacy lakehouse configurations, the use of shell escapes to modify hidden system-level files bypasses standard administrative interfaces and can impact platform stability.
Privilege Escalation and Configuration Tampering: The recommendation to write data to a directory associated with a service user (/home/trusted-service-user/) represents a potential privilege escalation concern. Instructions that encourage modifying service-level configuration files via user-level notebook cells can lead to unauthorized changes in environment behavior and should be replaced with supported configuration APIs where possible.
Indirect Prompt Injection Surface: The skill is designed to orchestrate the ingestion and processing of data from external sources into the Fabric Bronze layer (e.g., landing CSV or Parquet files). This creates an attack surface for indirect prompt injection, as the agent processes untrusted external content.
Ingestion points: Data ingestion occurs at the Bronze layer, where external files are landed in the lakehouse Files/ directory.
Boundary markers: The instructions do not specify the use of delimiters or warnings to prevent the agent from interpreting instructions that might be embedded within the source data.
Capability inventory: The skill utilizes sensitive capabilities including REST API operations (az rest), shell command execution (!echo), and file system write operations.
Sanitization: There are no explicit instructions for the sanitization or validation of the content within the ingested data before it is processed by the transformation notebooks.

e2e-medallion-architecture