The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection by ingesting and processing untrusted data from group room conversations and stored semantic memories.
Ingestion points: Data enters the agent context through the searchConversations and executeMemoryTool functions in handler.ts.
Boundary markers: While the handler utilizes renderHit and compactMemory to structure results, there are no explicit delimiters or instructions provided to the agent to disregard commands embedded within the retrieved memories.
Capability inventory: The skill enables the agent to perform state-changing operations including remember, update_memory, and delete_memory.
Sanitization: Content undergoes basic whitespace normalization and truncation in compactMemory, but lacks semantic filtering to prevent instructions within the data from influencing the agent's logic.

memory-management