The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous commands to be executed via the Databricks CLI (databricks) to manage projects, branches, and databases. It also instructs the user to run a local script scripts/grant_lakebase_permissions.py using uv to configure database access.
[EXTERNAL_DOWNLOADS]: The instructions involve adding databricks-langchain[memory] to the project dependencies. This package is provided by Databricks and is necessary for the memory integration.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the setup of persistent storage for agent memory, including short-term conversation history and long-term user facts. This configuration creates a data ingestion surface where external data stored in the database could influence agent behavior during retrieval, representing a potential indirect injection vector that must be managed by the agent's application logic.
Ingestion points: Persistent PostgreSQL storage (specifically checkpoint and store tables defined in Lakebase).
Boundary markers: Not explicitly defined in this setup guide.
Capability inventory: The storage is intended for use with LangChain and OpenAI SDK memory components which affect agent context.
Sanitization: Not addressed in this infrastructure setup guide.
[PRIVILEGE_ESCALATION]: Step 6 details the process of granting PostgreSQL roles and permissions (such as CAN_CONNECT_AND_CREATE, USAGE, SELECT, and INSERT) to the application's service principal. While this involves privilege management, it is a standard and necessary procedure for the application to interact with its required database resources.

lakebase-setup