The Agent Skills Directory

[Dynamic Code Deployment]: The skill facilitates the creation and deployment of User-Defined Functions (UDFs) in SQL, JavaScript, and Python.
Context: While this involves runtime execution of logic, BigQuery executes these functions within managed, sandboxed environments (such as the V8 engine for JavaScript and isolated containers for Python) to maintain environment security.
[External Dependency Management]: The instructions describe how to include third-party Python packages from PyPI and JavaScript libraries hosted in Google Cloud Storage.
Context: This is a standard platform feature for extending UDF capabilities. The skill promotes best practices such as pinning package versions and ensuring Cloud Storage artifacts are accessed with appropriate IAM permissions (e.g., roles/storage.objectViewer).
[IAM and Privilege Management]: The skill provides procedures for configuring IAM roles, including granting permissions to service accounts and modifying project-level policies.
Context: These high-privilege operations are necessary for managing BigQuery infrastructure. The skill explicitly advises users to confirm these actions before execution and follows the principle of least privilege by using dedicated service accounts for external connections.
[Outbound Network Considerations]: Certain UDF configurations (Remote Functions and Python UDFs with WITH CONNECTION) are capable of making outbound network requests.
Context: The skill manages this capability through dedicated 'Connection' objects and provides implementation templates that include explicit timeouts and prescriptive exception handling to ensure stability and security.
[Data Privacy and Logging]: The skill includes a specific security directive to avoid logging raw data values within UDF code.
Context: This measure prevents sensitive data processed by the UDF from being exposed in Cloud Logging to personnel who may have logging access but lack permissions to view the underlying BigQuery data.
[Indirect Prompt Injection Surface]: Because UDFs process data from BigQuery tables that may originate from untrusted external sources, there is a potential for data to influence UDF logic or downstream API calls.
Context: The skill addresses this surface area by encouraging defensive programming techniques, such as strictly avoiding the logging of row-level data and implementing granular exception handling to prevent unexpected data from disrupting the execution flow. (Evidence: Ingestion at UDF parameters in SKILL.md; Boundary markers provided by SQL syntax; Capabilities include network calls and infrastructure commands; Sanitization includes guidelines against logging raw data).

bigquery-udfs