Context

The multi-agent substrate currently relies on a per-process singleton (GraphService) for its in-memory cache. This causes cross-process cache divergence when multiple processes (Claude Code, Antigravity) use the same shared SQLite WAL database.

The Problem

Process A seeds a node, but Process B's vicinityLoadedNodes cache has no knowledge of it.

The Architectural Reality

The current cache is in-memory per-process.

The Fix

Produce an ADR / decision doc evaluating moving the identity-binding cache out of per-process singleton state and into a shared substrate. Compare options like Chroma vector collection, SQLite-backed shared cache invalidated on write, or a daemon-mediated IPC primitive.

Acceptance Criteria

• Produce an ADR decision artifact comparing shared caching options.
• Select a direction to implement.

Out of Scope

• Implementation of the chosen substrate (handled in subsequent ticket).

Avoided Traps

• Do not assume a within-process retry loop fixes systemic cross-process cache divergence.

Related

• Parent Epic: #10186

Handoff Retrieval Hints

• query_raw_memories(query="cross-harness MCP singleton cache divergence")