fix(openclaw): avoid memory retain recursion (#362)

hindsight-docs/docs/sdks/integrations/openclaw.md

@@ -59,6 +59,8 @@ The plugin will automatically:
 
 **Auto-Recall:** Before each agent response, relevant memories are automatically injected into the context (up to 1024 tokens). The agent uses past context without needing to call tools.
 
+**Feedback Loop Prevention:** The plugin automatically strips injected memory tags (`<hindsight_memories>`) before storing conversations. This prevents recalled memories from being re-extracted as new facts, which would cause exponential memory growth and duplicate entries.
+
 Traditional memory systems give agents a `search_memory` tool - but models don't use it consistently. Auto-recall solves this by injecting memories automatically before every turn.
 
 ## Configuration

hindsight-integrations/openclaw/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@vectorize-io/hindsight-openclaw",
-  "version": "0.4.7",
+  "version": "0.4.10",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@vectorize-io/hindsight-openclaw",
-      "version": "0.4.7",
+      "version": "0.4.10",
       "license": "MIT",
       "dependencies": {
         "node-fetch": "^3.3.2"

hindsight-integrations/openclaw/src/index.ts

@@ -702,6 +702,12 @@ User message: ${prompt}
                 .join('\n');
             }
 
+            // Strip plugin-injected memory tags to prevent feedback loop
+            // Remove <hindsight_memories> blocks injected during before_agent_start
+            content = content.replace(/<hindsight_memories>[\s\S]*?<\/hindsight_memories>/g, '');
+            // Remove any <relevant_memories> blocks (legacy/alternative format)
+            content = content.replace(/<relevant_memories>[\s\S]*?<\/relevant_memories>/g, '');
+
             return `[role: ${role}]\n${content}\n[${role}:end]`;
           })
           .join('\n\n');