The Agent Skills Directory

[SAFE]: The skill serves as a comprehensive developer guide for the Databricks Apps platform. It instructs users on scaffolding apps using official CLI tools (databricks apps init) and leveraging the AppKit SDK.
[EXTERNAL_DOWNLOADS]: The skill references the installation of official vendor packages (@databricks/appkit, @databricks/lakebase) and well-known development tools (zod, playwright, vitest) from standard package registries. These are documented neutrally as part of the standard development workflow.
[INDIRECT_PROMPT_INJECTION]: The skill addresses potential injection risks in AI-enabled features (like the agents and genie plugins). It provides explicit guidance on using Zod for schema validation and strictly parameterized SQL queries to mitigate these risks.
Ingestion points: Data entering through Express routes (server/server.ts), Unity Catalog Volumes, and chat interfaces.
Boundary markers: Recommendations for explicit data boundaries and 'ignore embedded instructions' patterns in agent logic.
Capability inventory: File system access (via files plugin), database operations (via lakebase plugin), and job execution (via jobs plugin).
Sanitization: Strong emphasis on using Zod for input parsing and Positional/Named parameters in SQL to prevent command and SQL injection.
[COMMAND_EXECUTION]: Instructions involve using the Databricks CLI for platform operations (e.g., databricks apps deploy). These are standard administrative tasks for the targeted environment and do not involve arbitrary or hidden command execution.
[CREDENTIALS_SAFE]: The skill correctly advises using environment variables and Databricks Secrets for sensitive configuration, following industry best practices for secret management.

databricks-apps