Bug: active-memory plugin crashes gateway when mx-claw plugin is enabled

Environment

• OpenClaw: 2026.4.10 (also reproducible on 2026.4.11-beta.1)
• Channel: webchat (direct)
• Plugins active: mx-claw, memory-core, feishu, openclaw-weixin, lossless-claw, browser, minimax, openclaw-mcp-bridge, duckduckgo
• active-memory: enabled with default config (agents: ["main"], allowedChatTypes: ["direct"], queryMode: "recent", timeoutMs: 10000)

Problem

Enabling the active-memory plugin causes the gateway to become completely unresponsive (hard crash). The agent stops responding to any messages. Deleting the active-memory configuration from plugins.entries immediately restores normal operation.

The crash is not caused by the timeoutMs being too short — setting it to 30000ms does not prevent the crash. The issue occurs immediately on the first message after the plugin is activated.

Root Cause Analysis

active-memory spawns a dedicated memory sub-agent (agent:main:main:active-memory:xxx) right before each main reply (as documented in the 4.10 changelog: "gives OpenClaw a dedicated memory sub-agent right before the main reply").

When mx-claw plugin is enabled, mx-claw registers a before_agent_start hook that intercepts all agent startup events, including the active-memory sub-agent session. The mx-claw hook checks for a valid channelId on the session, but the active-memory sub-agent session does not carry the original webchat channelId, causing mx-claw to reject the session.

Error observed in logs:

sessionKey=agent:main:main:active-memory:xxx does not have a valid channelId or channelId does not belong to mx-claw

This creates a conflict between two plugins both trying to manage session lifecycle hooks.

Impact

• Gateway becomes completely unresponsive
• No recovery without manually editing the JSON config to remove active-memory
• Cannot use active-memory and mx-claw together

Expected Behavior

active-memory should work alongside mx-claw without crashing the gateway. The sub-agent sessions created by active-memory should either:

1. Not trigger mx-claw's before_agent_start hook, or
2. Carry sufficient context (channelId) to pass mx-claw's validation

Configuration

"plugins": {
  "allow": ["openclaw-weixin", "feishu", "lossless-claw", "mx-claw", "browser", "minimax", "openclaw-mcp-bridge", "duckduckgo", "memory-core", "active-memory"],
  "entries": {
    "mx-claw": {
      "enabled": true,
      "config": {
        "auth": { "token": "em_..." },
        "serverUrl": "https://mx-claw-server.eastmoney.com/chat"
      }
    },
    "active-memory": {
      "enabled": true,
      "config": {
        "agents": ["main"],
        "allowedChatTypes": ["direct"],
        "queryMode": "recent",
        "timeoutMs": 10000
      }
    }
  }
}

Workaround

Remove active-memory from plugins.entries and plugins.allow to prevent the crash. mx-claw works normally without active-memory.

Tags

plugin, active-memory, mx-claw, conflict, crash