My agent and I have been working through several issues today when trying to implement your plugins. The biggest being a continuity bug causing the gateway to crash.
Environment
Linux x64, containerized (Unraid environment)
Node.js 22.x
OpenClaw gateway running with plugin allowlist + explicit plugins.load.paths
Continuity: kernel-level Node segfault (native crash)
Symptoms
With Continuity enabled, the gateway intermittently segfaults (kernel log shows segfault ... in node).
User-visible effect: sometimes the web UI shows the response but Discord never receives it (gateway dies mid-send).
Disabling Continuity stops the segfaults immediately.
Likely crash surface
Continuity uses native components:
better-sqlite3 native addon (better_sqlite3.node)
sqlite-vec which loads a native SQLite extension (vec0.so) via db.loadExtension(getLoadablePath())
This looks like a native compatibility / unsafe invocation problem, not a JS exception. A config conflict (memory settings, etc.) would not crash node at the kernel level.
Additional signal: empty-query search calls
We saw Continuity logging search calls like:
Search: intent=false, len=0, query=""
If Continuity is calling into sqlite-vec / vector query paths with an empty query (or similar edge-case), that might be triggering undefined behavior in native code.
Repro steps (Continuity segfault)
- Enable Continuity in OpenClaw plugin config (plugins.entries.continuity.enabled = true)
- Restart gateway
- Ask multiple “old memory” / archive search type questions
- Observe intermittent gateway crashes: kernel log shows segfault in node
- Disable continuity → crashes stop
Side note: Documentation makes mention of a "Recalled Memories" section in AGENTS.md, but details of what should be added are not specified anywhere
My agent and I have been working through several issues today when trying to implement your plugins. The biggest being a continuity bug causing the gateway to crash.
Environment
Linux x64, containerized (Unraid environment)
Node.js 22.x
OpenClaw gateway running with plugin allowlist + explicit plugins.load.paths
Continuity: kernel-level Node segfault (native crash)
Symptoms
With Continuity enabled, the gateway intermittently segfaults (kernel log shows segfault ... in node).
User-visible effect: sometimes the web UI shows the response but Discord never receives it (gateway dies mid-send).
Disabling Continuity stops the segfaults immediately.
Likely crash surface
Continuity uses native components:
better-sqlite3 native addon (better_sqlite3.node)
sqlite-vec which loads a native SQLite extension (vec0.so) via db.loadExtension(getLoadablePath())
This looks like a native compatibility / unsafe invocation problem, not a JS exception. A config conflict (memory settings, etc.) would not crash node at the kernel level.
Additional signal: empty-query search calls
We saw Continuity logging search calls like:
Search: intent=false, len=0, query=""
If Continuity is calling into sqlite-vec / vector query paths with an empty query (or similar edge-case), that might be triggering undefined behavior in native code.
Repro steps (Continuity segfault)
Side note: Documentation makes mention of a "Recalled Memories" section in AGENTS.md, but details of what should be added are not specified anywhere