memory-query

To query institutional memory before acting, apply a pattern using presence checks and ToolSearch to fetch past patterns from ruvector's vector store, then filter and deduplicate results to inform execution.

If ruvector or the MCP server is unavailable, memory queries degrade gracefully by performing presence checks first, allowing the agent to continue execution safely without relying on the missing institutional memory store.

An agent safely recalls past patterns by using ToolSearch combined with a safe reflexion XML format to surface findings, preventing leakage or execution of any embedded instructions stored in the vector memory.

You need to consult institutional memory during agent development when new commands or agents must learn from prior knowledge, ensuring they reference past solutions and findings before executing their assigned tasks.

Yes, you can use ToolSearch to retrieve past solutions safely by wrapping results in a reflexion XML format, which surfaces findings for the agent to read while preventing the execution of any stored instructions.

The best way to avoid duplicate findings when querying vector memory is to apply a workflow with explicit result filtering and deduplication steps after the ToolSearch fetches patterns from the institutional store.