The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The 'Metric View Advisor' workflow ingests untrusted data from multiple sources including SQL files, Genie Agent instructions, and AI/BI Dashboard queries to generate metric recommendations.
Ingestion Points: Data is read from user-provided .sql, .csv, and .yaml files, as well as fetched from Genie Agents (genie get-space) and Dashboards (lakeview get) in metric-view-advisor.md.
Boundary Markers: The skill generates a suggestions.yaml file and requires the user to 'Approve as-is' or 'Edit the file' before proceeding to creation.
Capability Inventory: The skill can execute SQL via databricks experimental aitools tools query, write files to the local workspace, and deploy DDL via the Databricks Statement Execution API (databricks api post /api/2.0/sql/statements/).
Sanitization: The workflow includes mandatory 'Review gates' (Step 4) and 'Pre-deploy checklists' (Step 6) that require explicit human approval of the final YAML/SQL definition before execution.
[COMMAND_EXECUTION]: The skill instructs the agent to use the Databricks CLI for workspace operations and deployment.
Evidence: Use of databricks auth, databricks bundle, and databricks api post for deploying metric view definitions as documented in SKILL.md and metric-view-advisor.md.
[DYNAMIC_EXECUTION]: The skill dynamically generates YAML-based metric view definitions at runtime based on the analysis of external assets.
Evidence: The 'Advisor' synthesizes metadata from input sources into new CREATE OR REPLACE VIEW statements which are then submitted to the platform's SQL engine, as detailed in the Step 4 and Step 6 workflows of metric-view-advisor.md.

databricks-metric-views