The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions utilize npx @databricks/appkit to generate types and access documentation. This involves downloading and executing tools from Databricks, which is recognized as a well-known technology service provider.\n- [COMMAND_EXECUTION]: The skill executes databricks CLI commands and local Python scripts (validate_contract.py) to validate report configurations and submit SQL statements to Databricks warehouses. These actions are restricted to the intended functionality of managing report contracts.\n- [PROMPT_INJECTION]: The skill processes user-provided SQL and YAML configuration files within the reports/ directory, which represents an indirect prompt injection surface. This risk is effectively mitigated by the included scripts/validate_contract.py utility, which performs static analysis on the SQL to enforce security constraints.\n
Ingestion points: reports/<name>/report.yaml and reports/<name>/queries/*.sql are loaded and parsed.\n
Boundary markers: While no specific LLM delimiters are used for the file contents, the validation script acts as a mandatory gatekeeper before execution.\n
Capability inventory: The skill can execute SQL via the Databricks Statement Execution API (/api/2.0/sql/statements/).\n
Sanitization: The validate_contract.py script identifies and rejects SQL containing DDL/DML keywords, string interpolation, SELECT *, or unqualified relations to prevent unauthorized data access or modifications.

governed-report-contract