Feature: cross-workflow agent memory — persistent verified memory for AI Agent nodes

I maintain [cognicore-env]( (open-source, MIT) — an agent memory layer that runs as an MCP server with no vector database or embedding server. N8n agent workflows can consume it today with zero changes; this issue is about making that a first-class story. The gap: n8n's AI Agent nodes have memory nodes (chat history per workflow), but knowledge does not cross workflows or executions. An agent in the lead-qualification workflow that learned what makes a lead convert cannot hand that to the support-workflow agent; each workflow's agent re-learns from scratch across the same 1500+ integrations. Proposal: an MCP-based memory node story, or a community node wrapping our transfer layer. Memories carry proof (command, result, exit code, output hash) and chain of custody; importing workflows receive them as observed-until-reverified; trust decays across unverified hops. Release rules prevent memories recorded in a private-repo-triggered workflow from surfacing in a shared-channel …