docs: update memory tool list, CLI usage, and config file reading logic (#4861)

Co-authored-by: Livia Ellen <liviaellen@msn.com>

openclaw/filtering.ts

@@ -14,9 +14,41 @@ const NOISE_MESSAGE_PATTERNS: RegExp[] = [
   /^(HEARTBEAT_OK|NO_REPLY)$/i,
   /^Current time:.*\d{4}/,
   /^Pre-compaction memory flush/i,
-  /^(ok|yes|no|sir|sure|thanks|done|good|nice|cool|got it|it's on|continue)$/i,
+  /^(ok|yes|no|sir|sure|thanks|done|good|nice|cool|got it|it's on|continue|alright|okay|yep|nope|uh-huh|mm-hmm|hmm)$/i,
   /^System: \[.*\] (Slack message edited|Gateway restart|Exec (failed|completed))/,
   /^System: \[.*\] ⚠️ Post-Compaction Audit:/,
+  // JSON-only messages (tool results, metadata)
+  /^[\s]*\{[\s\S]*\}[\s]*$/,
+  /^[\s]*\[[\s\S]*\][\s]*$/,
+  // Empty or whitespace-only after trimming
+  /^[\s\n\r]*$/,
+  // Technical noise patterns
+  /^(Error|Warning|Info|Debug):/i,
+  /^(Loading|Loaded|Fetching|Fetched|Processing|Processed)\b/i,
+  /^\[[\d:T\-\.Z]+\]/,  // Timestamps like [2024-01-01T12:00:00.000Z]
+  /^(SUCCESS|FAILURE|PENDING|COMPLETED|FAILED)$/i,
+  // Tool/function call noise
+  /^(Calling|Called|Invoking|Invoked|Executing|Executed)\s+(function|tool|method)/i,
+  /^Tool (call|result|output):/i,
+  // Single emoji or very short messages
+  /^[\p{Emoji}\s]{1,5}$/u,
+];
+
+/** Patterns for session-specific technical content that should not be stored as memories. */
+const SESSION_SPECIFIC_PATTERNS: RegExp[] = [
+  // Tool availability discussions
+  /tools?\s+(are|is)\s+(not\s+)?(exposed|available|accessible)/i,
+  /plugin\s+(does not|doesn't)\s+expose/i,
+  /I\s+(do not|don't)\s+(currently\s+)?see\s+.*tools?\s+exposed/i,
+  /memory_(search|get|add|update|delete|list)\s+(tool|is|are)/i,
+  // Session-specific capability statements
+  /in\s+this\s+session/i,
+  /my\s+(live\s+)?callable\s+tool\s+registry/i,
+  /tools?\s+I\s+(have|currently have)\s+access\s+to/i,
+  // Plugin/capability status statements
+  /openclaw-mem0\s+plugin/i,
+  /memory\s+wiki.*capability/i,
+  /workspace\s+memory\s+files/i,
 ];
 
 /** Content fragments that should be stripped from otherwise-valid messages. */
@@ -50,6 +82,31 @@ const NOISE_CONTENT_PATTERNS: Array<{ pattern: RegExp; replacement: string }> =
         /Replied message \(untrusted, for context\):\s*```json[\s\S]*?```/g,
       replacement: "",
     },
+    // Strip embedded JSON blocks that might contain metadata
+    {
+      pattern: /```json\s*\{[\s\S]*?\}\s*```/g,
+      replacement: "",
+    },
+    // Strip code blocks that are just tool outputs
+    {
+      pattern: /```(?:text|output|result|log)\s*[\s\S]*?```/gi,
+      replacement: "",
+    },
+    // Strip inline tool call IDs
+    {
+      pattern: /\[tool_call_id:[^\]]+\]/g,
+      replacement: "",
+    },
+    // Strip memory IDs from responses
+    {
+      pattern: /\(id:\s*[a-f0-9-]+\)/gi,
+      replacement: "",
+    },
+    // Strip session/run IDs
+    {
+      pattern: /(?:session|run|agent)[_-]?(?:id|key)?:\s*[a-zA-Z0-9_:-]+/gi,
+      replacement: "",
+    },
   ];
 
 const MAX_MESSAGE_LENGTH = 2000;
@@ -81,6 +138,20 @@ export function isNoiseMessage(content: string): boolean {
   return NOISE_MESSAGE_PATTERNS.some((p) => p.test(trimmed));
 }
 
+/**
+ * Check whether a message contains session-specific technical content
+ * that should not be stored (tool availability discussions, plugin
+ * capability statements, etc.). These are facts about the current session
+ * that have no value in future sessions.
+ */
+export function isSessionSpecificContent(content: string): boolean {
+  const trimmed = content.trim();
+  if (!trimmed) return false;
+  // Check if multiple session-specific patterns match (more confident filtering)
+  const matches = SESSION_SPECIFIC_PATTERNS.filter((p) => p.test(trimmed));
+  return matches.length >= 2;
+}
+
 /**
  * Check whether an assistant message is a generic acknowledgment with no
  * extractable facts (e.g. "I see you've shared an update. How can I help?").
@@ -120,7 +191,7 @@ function truncateMessage(content: string): string {
 
 /**
  * Full pre-extraction pipeline: drop noise messages, strip noise fragments,
- * and truncate remaining messages to a reasonable length.
+ * filter session-specific content, and truncate remaining messages.
  */
 export function filterMessagesForExtraction(
   messages: Array<{ role: string; content: string }>,
@@ -131,6 +202,8 @@ export function filterMessagesForExtraction(
     // Drop generic assistant acknowledgments that contain no facts
     if (msg.role === "assistant" && isGenericAssistantMessage(msg.content))
       continue;
+    // Drop session-specific technical content (tool availability, plugin capabilities)
+    if (isSessionSpecificContent(msg.content)) continue;
     const cleaned = stripNoiseFromContent(msg.content);
     if (!cleaned) continue;
     filtered.push({ role: msg.role, content: truncateMessage(cleaned) });