Don't store system / agent summaries.

internal/worker/sdk/processor.go

@@ -642,25 +642,57 @@ Your response:`,
 // rather than actual user work. These summaries should be filtered out.
 func isSelfReferentialSummary(summary *models.ParsedSummary) bool {
 	// Combine all summary fields for checking
-	content := strings.ToLower(summary.Request + " " + summary.Completed + " " + summary.Learned + " " + summary.NextSteps)
+	content := strings.ToLower(summary.Request + " " + summary.Completed + " " + summary.Learned + " " + summary.NextSteps + " " + summary.Investigated + " " + summary.Notes)
 
 	// Indicators that the summary is about the memory agent, not user work
 	selfReferentialPhrases := []string{
+		// Agent references
 		"memory extraction",
 		"memory agent",
+		"extraction agent",
 		"hook execution",
 		"hook mechanism",
+		// Session meta-state
 		"session initialization",
 		"session setup",
+		"session has just started",
+		"session just started",
 		"agent initialization",
 		"no technical learnings",
 		"no code or project work",
+		// Waiting states
 		"waiting for the user",
 		"waiting for user",
 		"awaiting actual",
 		"awaiting claude code",
+		"awaiting tool",
+		"awaiting user",
+		// Meta checkpoint references
 		"progress checkpoint",
 		"checkpoint request",
+		// Common no-work phrases
+		"no work has been completed",
+		"no work completed",
+		"no work done",
+		"no actual work",
+		"nothing has been completed",
+		"nothing completed",
+		// Role/guideline parroting
+		"role definition",
+		"operational guidelines",
+		"providing role",
+		"providing guidelines",
+		// System prompt echoes
+		"extract meaningful observations",
+		"meaningful learnings",
+		"analyze tool executions",
+		"observations for future sessions",
+		// Empty session indicators
+		"empty session",
+		"no substantive work",
+		"no meaningful work",
+		"just beginning",
+		"just begun",
 	}
 
 	matchCount := 0
@@ -685,20 +717,33 @@ func hasMeaningfulContent(assistantMsg string) bool {
 
 	lowerMsg := strings.ToLower(assistantMsg)
 
-	// Skip messages that are primarily about system/hook status
+	// Skip messages that are primarily about system/hook status or meta-instructions
 	skipIndicators := []string{
+		// System/hook markers
 		"hook success",
 		"callback hook",
 		"session start",
 		"sessionstart",
 		"system-reminder",
+		// Agent self-references
 		"memory extraction agent",
 		"memory agent",
+		"extraction agent",
+		// No-work indicators
 		"no technical learnings",
 		"waiting for",
 		"waiting to",
 		"no code or project work",
 		"no substantive",
+		"no work has been completed",
+		"no work done",
+		"awaiting tool",
+		"awaiting user",
+		// Meta-instruction echoes
+		"role definition",
+		"operational guidelines",
+		"analyze tool executions",
+		"extract meaningful observations",
 	}
 
 	skipCount := 0

internal/worker/sdk/processor_test.go

@@ -0,0 +1,126 @@
+package sdk
+
+import (
+	"testing"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+func TestIsSelfReferentialSummary(t *testing.T) {
+	tests := []struct {
+		name     string
+		summary  *models.ParsedSummary
+		expected bool
+	}{
+		{
+			name: "meta summary about memory agent role",
+			summary: &models.ParsedSummary{
+				Request:   "Memory extraction agent role - analyze tool executions and extract meaningful observations for future sessions",
+				Completed: "No work has been completed yet. The session has just started with the user providing role definition and operational guidelines.",
+				Learned:   "The system expects observations to be created from meaningful learnings during Claude Code sessions, with focus on decisions, bugs fixed, patterns discovered, project structure changes, and code modifications.",
+				NextSteps: "Awaiting tool executions or user requests that contain actual work performed in a Claude Code session.",
+			},
+			expected: true,
+		},
+		{
+			name: "legitimate summary about code changes",
+			summary: &models.ParsedSummary{
+				Request:   "Fix authentication bug in login handler",
+				Completed: "Updated the auth middleware to properly validate JWT tokens and fixed the session expiry check.",
+				Learned:   "The JWT library requires explicit algorithm validation to prevent token substitution attacks.",
+				NextSteps: "Add unit tests for the authentication flow.",
+			},
+			expected: false,
+		},
+		{
+			name: "awaiting user summary",
+			summary: &models.ParsedSummary{
+				Request:   "Session initialization",
+				Completed: "No work completed yet.",
+				Learned:   "Awaiting user input to begin work.",
+				NextSteps: "Waiting for the user to provide instructions.",
+			},
+			expected: true,
+		},
+		{
+			name: "summary about refactoring",
+			summary: &models.ParsedSummary{
+				Request:   "Refactor database connection pooling",
+				Completed: "Implemented connection pooling using pgxpool with max 10 connections.",
+				Learned:   "pgxpool automatically handles connection reuse and health checks.",
+				NextSteps: "Run benchmarks to verify performance improvement.",
+			},
+			expected: false,
+		},
+		{
+			name: "meta summary with extraction agent mention",
+			summary: &models.ParsedSummary{
+				Request:   "Extraction agent initialization",
+				Completed: "No substantive work has been done.",
+				Learned:   "The memory extraction agent analyzes tool executions.",
+				NextSteps: "Awaiting tool results to extract observations.",
+			},
+			expected: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := isSelfReferentialSummary(tt.summary)
+			if result != tt.expected {
+				t.Errorf("isSelfReferentialSummary() = %v, want %v", result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestHasMeaningfulContent(t *testing.T) {
+	tests := []struct {
+		name     string
+		content  string
+		expected bool
+	}{
+		{
+			name:     "empty content",
+			content:  "",
+			expected: false,
+		},
+		{
+			name:     "too short content",
+			content:  "Hello world",
+			expected: false,
+		},
+		{
+			name: "meta content about memory agent",
+			content: `This is the memory extraction agent role definition.
+The system expects you to analyze tool executions and extract meaningful observations.
+No work has been completed yet. Awaiting tool results from the user's session.`,
+			expected: false,
+		},
+		{
+			name: "legitimate code discussion",
+			content: `I've updated the handler.go file to fix the authentication bug.
+The function validateToken() was not checking token expiry correctly.
+I've added a check for exp claim and implemented proper error handling.
+The changes have been tested and the build passes.`,
+			expected: true,
+		},
+		{
+			name: "hook status messages",
+			content: `SessionStart:Callback hook success: Success
+The memory agent is waiting for user input.
+System-reminder about available tools.
+No substantive work performed yet.`,
+			expected: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := hasMeaningfulContent(tt.content)
+			if result != tt.expected {
+				t.Errorf("hasMeaningfulContent() = %v, want %v", result, tt.expected)
+			}
+		})
+	}
+}

marketplace.json

@@ -13,7 +13,7 @@
     {
       "name": "claude-mnemonic",
       "description": "Persistent memory system for Claude Code - Go implementation with SQLite and ChromaDB vector search",
-      "version": "0.6.4",
+      "version": "0.6.8",
       "author": {
         "name": "lukaszraczylo",
         "url": "https://github.com/lukaszraczylo"
@@ -23,25 +23,25 @@
       "license": "MIT",
       "homepage": "https://github.com/lukaszraczylo/claude-mnemonic",
       "releases": {
-        "latest": "0.6.4",
+        "latest": "0.6.8",
         "versions": {
-          "0.6.4": {
+          "0.6.8": {
             "releaseDate": "2025-12-16",
             "downloads": {
               "darwin-arm64": {
-                "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/v0.6.4/claude-mnemonic_0.6.4_darwin_arm64.tar.gz",
+                "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/v0.6.8/claude-mnemonic_0.6.8_darwin_arm64.tar.gz",
                 "format": "tar.gz",
-                "sha256": "5bb082901e8b4f4751d011675a235a5ed25403204732e09842b5ae75136f65ae"
+                "sha256": "1a796f9e47ee27e2d2c9425e1d1dea886771c9c3f0d4905dc414b5c0763665c6"
               },
               "linux-amd64": {
-                "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/v0.6.4/claude-mnemonic_0.6.4_linux_amd64.tar.gz",
+                "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/v0.6.8/claude-mnemonic_0.6.8_linux_amd64.tar.gz",
                 "format": "tar.gz",
-                "sha256": "dee774fbc39e78ee6ff3836403c2901471dd0d417f62e214088f6b8285b3b23d"
+                "sha256": "52221d0e0f10d312abd1d521840ac0ac8558145795b8fb63f60a411dd2331637"
               },
               "windows-amd64": {
-                "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/v0.6.4/claude-mnemonic_0.6.4_windows_amd64.zip",
+                "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/v0.6.8/claude-mnemonic_0.6.8_windows_amd64.zip",
                 "format": "zip",
-                "sha256": "e3973bc1107f9cc92cf95697c62138b4cab4d0d4e00f836dfc4592ca84bbdab7"
+                "sha256": "08af5820a48bda13d791c21d6e8cb47b0ddfbba101be47c0b8f1c03966749167"
               }
             }
           }

ui/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "claude-mnemonic-dashboard",
-  "version": "v0.6.5-25-g372942e-dirty",
+  "version": "v0.6.4-2-gb396d05-dirty",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "claude-mnemonic-dashboard",
-      "version": "v0.6.5-25-g372942e-dirty",
+      "version": "v0.6.4-2-gb396d05-dirty",
       "dependencies": {
         "vue": "^3.5.13"
       },

ui/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-mnemonic-dashboard",
-  "version": "v0.6.5-25-g372942e-dirty",
+  "version": "v0.6.4-2-gb396d05-dirty",
   "private": true,
   "type": "module",
   "scripts": {