The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the databricks-langchain[memory] package from official repositories.
[DATA_EXFILTRATION]: Facilitates the persistence of conversation history and user memories to a Databricks Lakebase instance. Data access is controlled via authenticated request contexts.
[COMMAND_EXECUTION]: Provides developer instructions for table initialization and application deployment using standard CLI tools like uv and databricks.
[PROMPT_INJECTION]: The save_user_memory tool (examples/memory_tools.py) represents a surface for indirect prompt injection as it stores model-generated content.
Ingestion points: data enters through the memory_data_json parameter in the save_user_memory tool.
Boundary markers: no explicit delimiters or safety instructions are embedded in the tool code to mitigate injection from the stored data.
Capability inventory: utilizes store.aput, store.asearch, and store.adelete to interact with Lakebase storage.
Sanitization: the implementation includes validation using json.loads() to ensure the input is a valid JSON dictionary before it is persisted.

agent-memory