The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a long-term memory system that ingests external data into the agent context, creating a surface for indirect prompt injection.\n- Ingestion points: The search_memory and get_memory tools in agent_server/utils_memory.py retrieve data from the Databricks memory store.\n- Boundary markers: MEMORY_INSTRUCTIONS in agent_server/agent.py provide clear delimiters and warnings, instructing the agent to treat memory as untrusted data.\n- Capability inventory: The agent possesses tools for memory modification (save_memory, update_memory) and workspace interaction.\n- Sanitization: There is no programmatic sanitization of the retrieved content; the skill relies on the LLM's adherence to the provided instructions.\n- [EXTERNAL_DOWNLOADS]: The skill makes network requests to Databricks platform services via the official REST API for memory operations.\n- [COMMAND_EXECUTION]: Setup documentation in SKILL.md includes shell commands for administrative tasks, such as creating the memory store and configuring service principal permissions.

managed-memory