@vainplex/openclaw-membrane 0.3.3 → 0.3.4

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## [0.3.4] — 2026-03-04
+
+### Fixed
+- **Assistant messages now captured in Membrane.** Added `agent_end` hook workaround for `message_sent` not firing from gateway delivery pipeline in OpenClaw v2026.3.x. Only captures the last assistant message per turn (avoids flooding with entire conversation history).
+- **Removed `after_tool_call` hook.** Tool calls are operational logs (~95% of Membrane volume) and drown out actual conversations. Tool data is already captured in the NATS event store.
+
 ## [0.3.0] — 2026-02-23
 
 ### Added

package/README.md

@@ -1,5 +1,8 @@
 # @vainplex/openclaw-membrane
 
+[![npm](https://img.shields.io/npm/v/@vainplex/openclaw-membrane)](https://www.npmjs.com/package/@vainplex/openclaw-membrane)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
 Membrane gRPC bridge for OpenClaw — persistent episodic memory via [GustyCube/membrane](https://github.com/GustyCube/membrane) sidecar.
 
 **What it does:** Every conversation, tool call, and decision flows into Membrane's biological memory model. Memories decay over time. Frequently accessed ones grow stronger. Your agent remembers what matters and forgets what doesn't.

package/dist/index.js

@@ -166,6 +166,34 @@ const plugin = {
         api.on('message_received', hookHandler('message_received'));
         api.on('message_sent', hookHandler('message_sent'));
         api.on('message_sending', hookHandler('message_sending'));
+        // Workaround: message_sent hook doesn't fire from gateway delivery pipeline
+        // in OpenClaw v2026.3.x. Use agent_end to capture assistant messages instead.
+        // Only captures the LAST assistant message to avoid flooding with entire history.
+        // TODO: Remove when upstream fixes plugin hook bridging. PR: openclaw/openclaw#XXX
+        api.on('agent_end', (event, ctx) => {
+            const e = event;
+            if (!e.success || !e.messages)
+                return;
+            const messages = e.messages;
+            // Find the LAST assistant message only (not entire history)
+            for (let i = messages.length - 1; i >= 0; i--) {
+                const msg = messages[i];
+                if (msg.role !== 'assistant')
+                    continue;
+                const content = typeof msg.content === 'string'
+                    ? msg.content
+                    : Array.isArray(msg.content)
+                        ? msg.content
+                            .filter((b) => b.type === 'text' && typeof b.text === 'string')
+                            .map((b) => b.text)
+                            .join('\n')
+                        : '';
+                if (!content || content.length < 10)
+                    break;
+                handleEvent({ type: 'message_sent', payload: { content }, context: ctx }, config, reliability, logger);
+                break; // Only the last one
+            }
+        });
         // after_tool_call removed — tool calls are operational logs, not memories.
         // They flood Membrane (~95% of volume) and drown out actual conversations.
         // Tool data is already captured in NATS event store.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vainplex/openclaw-membrane",
-  "version": "0.3.3",
+  "version": "0.3.4",
   "type": "module",
   "description": "Membrane gRPC bridge for OpenClaw — episodic memory ingestion, search tool, and auto-context injection via Membrane sidecar",
   "main": "dist/index.js",