Problem

OpenClaw loads the memory-openviking plugin in two separate contexts ([gateway] and [plugins]), each calling register() independently. Each call creates its own local clientPromise. However, start() only runs in one context — leaving the other context's promise permanently pending.

This causes the before_agent_start hook to hang forever on await getClient() whenever hookAgentId is set (i.e. always in a standard gateway deployment), because the hook runs in the context whose clientPromise was never resolved.

Symptoms:

• Bot receives message, typing indicator starts, never gets a response
• Gateway log shows running before_agent_start (1 handlers, sequential) with nothing after
• autoRecall and autoCapture appear completely broken

Fix

Introduce a module-level localClientPendingPromises map in client.ts:

• register() checks the map first. If an entry exists, it shares that promise instead of creating a new pending one.
• start() atomically claims the entry (deletes it from the map) — only the first start() call spawns the process and resolves the shared promise. Subsequent start() calls fall through to await getClient() which waits for the spawner.
• stop() clears the map entry to prevent stale promises persisting across hot-reloads.

Test plan

• Send a message via Telegram with autoRecall: true — confirm autoRecall done appears in logs within ~2 seconds
• Restart gateway — confirm subsequent messages still work (no stale promise from previous run)
• Confirm before_agent_start hook no longer hangs indefinitely

🤖 Generated with Claude Code