[BUG] agent_end hook blocked in OpenClaw — smartExtraction and memoryReflection silently fail

[jlin53882]

Bug Description

When memory-lancedb-pro is installed as a non-bundled plugin in OpenClaw, the agent_end typed hook is silently blocked at runtime. This causes smartExtraction and memoryReflection to never fire, with no error shown to the user.

Root Cause

OpenClaw's plugin registry enforces a security policy: non-bundled plugins that register conversation-accessing typed hooks (agent_end, llm_input, llm_output, before_agent_finalize) must explicitly opt-in with allowConversationAccess: true.

If not set, the hook is silently dropped — no error is thrown, no user-facing message appears. The only trace is this warning in the gateway log:

typed hook "agent_end" blocked because non-bundled plugins must set
plugins.entries.memory-lancedb-pro.hooks.allowConversationAccess=true

Reproduction Steps

1. Install memory-lancedb-pro as an external plugin in OpenClaw (not bundled)
2. Enable smartExtraction: true or sessionStrategy: "memoryReflection" in plugin config
3. Observe in the gateway log that the hooks are blocked with the message above
4. Confirm that no memory entries are created after conversations end

Expected Behavior

• smartExtraction should extract structured facts after each conversation turn
• memoryReflection should store session summaries to LanceDB
• Either: hooks should work out of the box, OR the user should see a clear error at startup (not a silent warning)

Workaround

Add hooks.allowConversationAccess: true to the plugin entry in openclaw.json:

"plugins": {
  "entries": {
    "memory-lancedb-pro": {
      "enabled": true,
      "hooks": {
        "allowConversationAccess": true
      },
      "config": {
        "sessionStrategy": "memoryReflection",
        "smartExtraction": true
      }
    }
  }
}

Suggested Fix (Priority Order)

1. Best: In the plugin's README / integration docs, clearly document the required config path (plugins.entries.memory-lancedb-pro.hooks.allowConversationAccess=true) so users know to add it
2. Better: At plugin startup, detect if it's running under OpenClaw (non-bundled) and push a user-facing warning via the plugin API when hooks are blocked
3. Good: Add a preflight check in onLoad / setup entry that validates typed hooks are not blocked

Environment

• OpenClaw version: 1.1.0-beta.10+
• memory-lancedb-pro version: latest
• Platform: Windows (WSL2)
• Plugin installation: external (non-bundled)

Gateway Log Sample

[registry] typed hook "agent_end" blocked because non-bundled plugins must set
  plugins.entries.memory-lancedb-pro.hooks.allowConversationAccess=true
[memory-lancedb-pro] runMemoryReflection skipped: agentEndHooksDisabled

[jlin53882]

openclaw 4.25 以上才需要
但是 4.23 會無法正常使用 smartExtraction

[zeeyuu222]

Confirming this on memory-lancedb-pro 1.1.0-beta.9 + OpenClaw 2026.4.24 → 2026.4.26.

The agent_end hook started being blocked silently from OpenClaw 2026.4.24 onward. In our case, auto-capture and smart extraction were dead for 4 days before we noticed — the only clue was this log line in gateway.err.log:

[gateway] [plugins] typed hook "agent_end" blocked because non-bundled plugins must set
plugins.entries.memory-lancedb-pro.hooks.allowConversationAccess=true

The log is easy to miss because:

1. It only appears in gateway.err.log, not in the main log
2. autoRecall (injection) continues to work fine, so the plugin looks healthy
3. No startup warning or health-check flag indicates that write-side hooks are disabled

Workaround: manually add to openclaw.json:

{
  "plugins": {
    "entries": {
      "memory-lancedb-pro": {
        "hooks": {
          "allowConversationAccess": true
        }
      }
    }
  }
}

Suggestion: The plugin should either:

• Declare the hook requirement in its manifest/package.json so OpenClaw can auto-grant or prompt during setup
• Or at minimum surface a clear startup warning when the required hook is blocked (e.g. "⚠️ auto-capture disabled: allowConversationAccess not set")

Environment: macOS 15 (Darwin 25.4.0 arm64), Node v24.15.0

[jlin53882]

Thank you, I have solved problem. zeeyuu222 ***@***.***> 於 2026年4月30日 週四 上午8:52 寫道：

…

*zeeyuu222* left a comment (CortexReach/memory-lancedb-pro#707) <#707 (comment)> Confirming this on *memory-lancedb-pro 1.1.0-beta.9* + *OpenClaw 2026.4.24 → 2026.4.26*. The agent_end hook started being blocked silently from OpenClaw 2026.4.24 onward. In our case, *auto-capture and smart extraction were dead for 4 days* before we noticed — the only clue was this log line in gateway.err.log: [gateway] [plugins] typed hook "agent_end" blocked because non-bundled plugins must set plugins.entries.memory-lancedb-pro.hooks.allowConversationAccess=true The log is easy to miss because: 1. It only appears in gateway.err.log, not in the main log 2. autoRecall (injection) continues to work fine, so the plugin *looks* healthy 3. No startup warning or health-check flag indicates that write-side hooks are disabled *Workaround*: manually add to openclaw.json: { "plugins": { "entries": { "memory-lancedb-pro": { "hooks": { "allowConversationAccess": true } } } } } *Suggestion*: The plugin should either: - Declare the hook requirement in its manifest/package.json so OpenClaw can auto-grant or prompt during setup - Or at minimum surface a clear startup warning when the required hook is blocked (e.g. "⚠️ auto-capture disabled: allowConversationAccess not set") Environment: macOS 15 (Darwin 25.4.0 arm64), Node v24.15.0 — Reply to this email directly, view it on GitHub <#707 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ASID5C4K7JHPTBDIM4RCOWD4YKPTBAVCNFSM6AAAAACYIH3J4OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DGNBYGY3DCNRSG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[jlin53882]

Plugin-level fix: declare allowConversationAccess in manifest

Instead of requiring every user to manually add the workaround to their openclaw.json, the plugin should declare this requirement in its own manifest (openclaw.plugin.json):

{
  "name": "memory-lancedb-pro",
  "version": "...",
  "hooks": {
    "allowConversationAccess": true
  }
}

This way OpenClaw can auto-grant the permission at plugin load time, and users don't need to touch their openclaw.json at all.

I have confirmed this works: adding hooks.allowConversationAccess: true to the plugin's own manifest eliminates the silent hook blocking with no side effects.

[jlin53882]

This needs to be decided by the maintainer @rwmjhb, it's not something I can decide. You can pull the latest commit and apply it along with PR727 to fix the issue, but there might be some bugs that weren't caught.

…

Message ID: ***@***.*** com>

[jlin53882]

It is currently advised against updating OpenClaw to versions above 4.23, as they contain numerous bugs. 林家豪 ***@***.***> 於 2026年4月30日 週四 下午10:09 寫道：

…

This needs to be decided by the maintainer @rwmjhb, it's not something I can decide. You can pull the latest commit and apply it along with PR727 to fix the issue, but there might be some bugs that weren't caught. > Message ID: ***@***.*** > com> >

[jlin53882]

@zeeyuu222 After my recent testing, smartexection/memory search works fine in openclaw version 4.22, but not in version 4.23.