claude-memory-agent 2.0.1 → 2.1.0

package/README.md

@@ -1,206 +1,206 @@
-# Claude Memory Agent
-
-A persistent semantic memory system for Claude Code sessions. Stores memories, tracks decisions, and provides anti-hallucination grounding context.
-
-## Features
-
-- **Semantic Memory Storage** - Store and retrieve memories using natural language
-- **Session Timeline** - Track decisions, actions, and observations
-- **Anti-Hallucination** - Grounding context injection to prevent contradictions
-- **Pattern Learning** - Store and reuse successful solution patterns
-- **Dashboard** - Web UI for browsing and managing memories
-- **Claude Code Integration** - Hooks for automatic session tracking
-
-## Quick Start
-
-### 1. Prerequisites (Install These First!)
-
-| Requirement | Download | Why Needed |
-|-------------|----------|------------|
-| **Python 3.9+** | https://python.org | Runs the agent |
-| **Ollama** | https://ollama.ai/download | Provides embeddings |
-| **Node.js 16+** | https://nodejs.org | For npm install |
-
-**After installing Ollama, run these commands:**
-
-```bash
-# Download the embedding model (required!)
-ollama pull nomic-embed-text
-
-# Start Ollama (runs in background)
-ollama serve
-```
-
-### 2. Installation
-
-#### Option A: Install via npm (Recommended)
-
-```bash
-# Install globally
-npm install -g claude-memory-agent
-
-# Run the setup wizard
-claude-memory-agent install
-
-# Start the agent
-claude-memory-agent start
-```
-
-#### Option B: Install via npx (No global install)
-
-```bash
-# Run directly without installing
-npx claude-memory-agent install
-npx claude-memory-agent start
-```
-
-#### Option C: Clone from GitHub
-
-```bash
-# Clone the repository
-git clone https://github.com/yourusername/claude-memory-agent.git
-cd claude-memory-agent
-
-# Run the installation wizard
-python install.py
-```
-
-The installer will:
-- Install Python dependencies
-- Create `.env` configuration file
-- Configure Claude Code MCP settings
-- Set up session tracking hooks
-- Create platform-specific startup scripts
-
-### 3. Start the Agent
-
-**Windows:**
-```bash
-# Option 1: Double-click start-memory-agent.vbs
-# Option 2: Use CLI
-memory-agent start
-```
-
-**Mac/Linux:**
-```bash
-./start-memory-agent.sh
-# or
-python memory-agent start
-```
-
-### 4. Open Dashboard
-
-```bash
-memory-agent dashboard
-# or open http://localhost:8102/dashboard
-```
-
-## CLI Commands
-
-### If installed via npm:
-
-```bash
-claude-memory-agent install    # Run setup wizard
-claude-memory-agent start      # Start agent in background
-claude-memory-agent stop       # Stop running agent
-claude-memory-agent status     # Check if agent is running
-claude-memory-agent dashboard  # Open web dashboard
-claude-memory-agent logs       # View recent log output
-```
-
-### If cloned from GitHub:
-
-```bash
-python memory-agent start      # Start agent in background
-python memory-agent stop       # Stop running agent
-python memory-agent status     # Check if agent is running
-python memory-agent dashboard  # Open web dashboard
-python memory-agent logs       # View recent log output
-python memory-agent install    # Re-run installation wizard
-```
-
-## Configuration
-
-Configuration is stored in `.env` (created during installation):
-
-```bash
-# Server
-PORT=8102
-HOST=0.0.0.0
-MEMORY_AGENT_URL=http://localhost:8102
-
-# Ollama / Embeddings
-OLLAMA_HOST=http://localhost:11434
-EMBEDDING_MODEL=nomic-embed-text
-
-# Database
-DATABASE_PATH=./memories.db
-USE_VECTOR_INDEX=true
-
-# Logging
-LOG_LEVEL=INFO
-```
-
-See `.env.example` for all available options.
-
-## MCP Tools
-
-The agent provides these tools to Claude Code:
-
-| Tool | Description |
-|------|-------------|
-| `memory_store` | Store a memory with semantic embedding |
-| `memory_search` | Search memories by natural language |
-| `memory_context` | Load session context and relevant memories |
-| `memory_search_patterns` | Find reusable solution patterns |
-| `memory_store_pattern` | Save a successful solution pattern |
-| `memory_get_project` | Get project-specific information |
-| `memory_store_project` | Store project configuration |
-| `memory_stats` | Get memory system statistics |
-| `memory_dashboard` | Open the web dashboard |
-
-## Project Structure
-
-```
-claude-memory-agent/
-├── main.py              # FastAPI server
-├── config.py            # Centralized configuration
-├── install.py           # Installation wizard
-├── memory-agent         # CLI tool
-├── services/            # Core services
-│   ├── database.py      # SQLite with connection pooling
-│   ├── embeddings.py    # Ollama integration
-│   ├── vector_index.py  # FAISS similarity search
-│   └── websocket.py     # Real-time updates
-├── skills/              # A2A skill implementations
-├── hooks/               # Claude Code hooks
-├── .env.example         # Configuration template
-└── .gitignore           # Excludes user-specific files
-```
-
-## Uninstall
-
-```bash
-python install.py --uninstall
-```
-
-This removes Claude Code integration but preserves your data.
-
-## Troubleshooting
-
-**Agent won't start:**
-- Check if Ollama is running: `ollama serve`
-- Check if port is in use: `netstat -an | grep 8102`
-- Check logs: `memory-agent logs`
-
-**No embeddings:**
-- Pull the model: `ollama pull nomic-embed-text`
-- Check Ollama host in `.env`
-
-**Hooks not working:**
-- Restart Claude Code after installation
-- Check `~/.claude/hooks.json` configuration
-
-## License
-
-MIT
+# Claude Memory Agent
+
+A persistent semantic memory system for Claude Code sessions. Stores memories, tracks decisions, and provides anti-hallucination grounding context.
+
+## Features
+
+- **Semantic Memory Storage** - Store and retrieve memories using natural language
+- **Session Timeline** - Track decisions, actions, and observations
+- **Anti-Hallucination** - Grounding context injection to prevent contradictions
+- **Pattern Learning** - Store and reuse successful solution patterns
+- **Dashboard** - Web UI for browsing and managing memories
+- **Claude Code Integration** - Hooks for automatic session tracking
+
+## Quick Start
+
+### 1. Prerequisites (Install These First!)
+
+| Requirement | Download | Why Needed |
+|-------------|----------|------------|
+| **Python 3.9+** | https://python.org | Runs the agent |
+| **Ollama** | https://ollama.ai/download | Provides embeddings |
+| **Node.js 16+** | https://nodejs.org | For npm install |
+
+**After installing Ollama, run these commands:**
+
+```bash
+# Download the embedding model (required!)
+ollama pull nomic-embed-text
+
+# Start Ollama (runs in background)
+ollama serve
+```
+
+### 2. Installation
+
+#### Option A: Install via npm (Recommended)
+
+```bash
+# Install globally
+npm install -g claude-memory-agent
+
+# Run the setup wizard
+claude-memory-agent install
+
+# Start the agent
+claude-memory-agent start
+```
+
+#### Option B: Install via npx (No global install)
+
+```bash
+# Run directly without installing
+npx claude-memory-agent install
+npx claude-memory-agent start
+```
+
+#### Option C: Clone from GitHub
+
+```bash
+# Clone the repository
+git clone https://github.com/yourusername/claude-memory-agent.git
+cd claude-memory-agent
+
+# Run the installation wizard
+python install.py
+```
+
+The installer will:
+- Install Python dependencies
+- Create `.env` configuration file
+- Configure Claude Code MCP settings
+- Set up session tracking hooks
+- Create platform-specific startup scripts
+
+### 3. Start the Agent
+
+**Windows:**
+```bash
+# Option 1: Double-click start-memory-agent.vbs
+# Option 2: Use CLI
+memory-agent start
+```
+
+**Mac/Linux:**
+```bash
+./start-memory-agent.sh
+# or
+python memory-agent start
+```
+
+### 4. Open Dashboard
+
+```bash
+memory-agent dashboard
+# or open http://localhost:8102/dashboard
+```
+
+## CLI Commands
+
+### If installed via npm:
+
+```bash
+claude-memory-agent install    # Run setup wizard
+claude-memory-agent start      # Start agent in background
+claude-memory-agent stop       # Stop running agent
+claude-memory-agent status     # Check if agent is running
+claude-memory-agent dashboard  # Open web dashboard
+claude-memory-agent logs       # View recent log output
+```
+
+### If cloned from GitHub:
+
+```bash
+python memory-agent start      # Start agent in background
+python memory-agent stop       # Stop running agent
+python memory-agent status     # Check if agent is running
+python memory-agent dashboard  # Open web dashboard
+python memory-agent logs       # View recent log output
+python memory-agent install    # Re-run installation wizard
+```
+
+## Configuration
+
+Configuration is stored in `.env` (created during installation):
+
+```bash
+# Server
+PORT=8102
+HOST=0.0.0.0
+MEMORY_AGENT_URL=http://localhost:8102
+
+# Ollama / Embeddings
+OLLAMA_HOST=http://localhost:11434
+EMBEDDING_MODEL=nomic-embed-text
+
+# Database
+DATABASE_PATH=./memories.db
+USE_VECTOR_INDEX=true
+
+# Logging
+LOG_LEVEL=INFO
+```
+
+See `.env.example` for all available options.
+
+## MCP Tools
+
+The agent provides these tools to Claude Code:
+
+| Tool | Description |
+|------|-------------|
+| `memory_store` | Store a memory with semantic embedding |
+| `memory_search` | Search memories by natural language |
+| `memory_context` | Load session context and relevant memories |
+| `memory_search_patterns` | Find reusable solution patterns |
+| `memory_store_pattern` | Save a successful solution pattern |
+| `memory_get_project` | Get project-specific information |
+| `memory_store_project` | Store project configuration |
+| `memory_stats` | Get memory system statistics |
+| `memory_dashboard` | Open the web dashboard |
+
+## Project Structure
+
+```
+claude-memory-agent/
+├── main.py              # FastAPI server
+├── config.py            # Centralized configuration
+├── install.py           # Installation wizard
+├── memory-agent         # CLI tool
+├── services/            # Core services
+│   ├── database.py      # SQLite with connection pooling
+│   ├── embeddings.py    # Ollama integration
+│   ├── vector_index.py  # FAISS similarity search
+│   └── websocket.py     # Real-time updates
+├── skills/              # A2A skill implementations
+├── hooks/               # Claude Code hooks
+├── .env.example         # Configuration template
+└── .gitignore           # Excludes user-specific files
+```
+
+## Uninstall
+
+```bash
+python install.py --uninstall
+```
+
+This removes Claude Code integration but preserves your data.
+
+## Troubleshooting
+
+**Agent won't start:**
+- Check if Ollama is running: `ollama serve`
+- Check if port is in use: `netstat -an | grep 8102`
+- Check logs: `memory-agent logs`
+
+**No embeddings:**
+- Pull the model: `ollama pull nomic-embed-text`
+- Check Ollama host in `.env`
+
+**Hooks not working:**
+- Restart Claude Code after installation
+- Check `~/.claude/hooks.json` configuration
+
+## License
+
+MIT

package/agent_card.py

@@ -248,6 +248,69 @@ AGENT_CARD = {
             "outputModes": ["text"]
         },
         # ============================================================
+        # SELF-CORRECTING CONFIDENCE SKILLS
+        # ============================================================
+        {
+            "id": "memory_worked",
+            "name": "Report Memory Worked",
+            "description": "Report that a memory's solution worked. Increases confidence by 0.15 and resets failure count. Use this when a retrieved solution successfully solved the problem.",
+            "tags": ["confidence", "feedback", "learning"],
+            "examples": [
+                "That solution from memory 42 worked perfectly",
+                "Report success for the HVP import fix"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "memory_failed",
+            "name": "Report Memory Failed",
+            "description": "Report that a memory's solution failed. Decreases confidence by 0.2. After 3 consecutive failures, memory is marked as unreliable and excluded from search.",
+            "tags": ["confidence", "feedback", "learning"],
+            "examples": [
+                "That solution from memory 42 didn't work",
+                "Report failure for the caching fix"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "get_reliability_stats",
+            "name": "Get Reliability Stats",
+            "description": "Get detailed reliability statistics for a memory including confidence, times worked/failed, success rate, and outcome history.",
+            "tags": ["confidence", "stats", "reliability"],
+            "examples": [
+                "What's the reliability of memory 42?",
+                "Show confidence stats for this solution"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "get_unreliable_memories",
+            "name": "Get Unreliable Memories",
+            "description": "Get all memories marked as unreliable (failure_count >= 3). These are excluded from search by default.",
+            "tags": ["confidence", "list", "unreliable"],
+            "examples": [
+                "Show unreliable memories",
+                "List failed solutions"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "reset_memory_reliability",
+            "name": "Reset Memory Reliability",
+            "description": "Reset a memory's reliability stats (admin function). Useful when a memory has been fixed or updated and should be given a fresh chance.",
+            "tags": ["confidence", "admin", "reset"],
+            "examples": [
+                "Give memory 42 a fresh start",
+                "Reset reliability for the fixed solution"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        # ============================================================
         # CLAUDE.MD MANAGEMENT SKILLS
         # ============================================================
         {
@@ -507,6 +570,129 @@ AGENT_CARD = {
             ],
             "inputModes": ["text"],
             "outputModes": ["text"]
+        },
+        # ============================================================
+        # CURATOR AGENT SKILLS (Graph Exploration & Maintenance)
+        # ============================================================
+        {
+            "id": "curator_explore",
+            "name": "Curator: Explore Graph",
+            "description": "Explore the memory graph from a starting node using BFS or DFS traversal. Returns connected nodes, edges, clusters, and insights about the subgraph.",
+            "tags": ["curator", "graph", "exploration"],
+            "examples": [
+                "Explore the graph from memory 42",
+                "Show connected memories 3 levels deep"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_find_duplicates",
+            "name": "Curator: Find Duplicates",
+            "description": "Find semantically similar (duplicate) memories using embedding similarity. Returns duplicate clusters with merge recommendations.",
+            "tags": ["curator", "duplicates", "cleanup"],
+            "examples": [
+                "Find duplicate memories",
+                "Show potential duplicates above 0.9 similarity"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_suggest_links",
+            "name": "Curator: Suggest Links",
+            "description": "Suggest missing relationships between memories based on semantic similarity. Returns potential links with confidence scores and relationship types.",
+            "tags": ["curator", "links", "relationships"],
+            "examples": [
+                "Suggest links for memory 42",
+                "Find potential relationships in this project"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_merge",
+            "name": "Curator: Merge Memories",
+            "description": "Merge duplicate memories into one. Transfers all relationships to the kept memory and archives the removed ones.",
+            "tags": ["curator", "merge", "cleanup"],
+            "examples": [
+                "Merge memory 43 into 42",
+                "Combine duplicate memories"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_get_summary",
+            "name": "Curator: Get Context Summary",
+            "description": "Generate a curated context summary for a query. Returns relevant memories organized by type with graph context and pending reviews.",
+            "tags": ["curator", "context", "summary"],
+            "examples": [
+                "Get curated context about authentication",
+                "Summarize what we know about the API"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_run_maintenance",
+            "name": "Curator: Run Maintenance",
+            "description": "Run curator maintenance tasks: dedup (find duplicates), orphans (find unconnected), links (suggest relationships), decay (reduce unused memory confidence), quality (score all memories).",
+            "tags": ["curator", "maintenance", "cleanup"],
+            "examples": [
+                "Run curator maintenance",
+                "Run dedup and orphan detection"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_get_report",
+            "name": "Curator: Get Report",
+            "description": "Get the latest curator maintenance report with findings, actions taken, and recommendations.",
+            "tags": ["curator", "report", "status"],
+            "examples": [
+                "Show latest curator report",
+                "What did the curator find?"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_get_status",
+            "name": "Curator: Get Status",
+            "description": "Get current curator agent status including total memories, relationships, orphan count, and connection ratio.",
+            "tags": ["curator", "status", "stats"],
+            "examples": [
+                "Show curator status",
+                "How is the memory graph doing?"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_score_quality",
+            "name": "Curator: Score Quality",
+            "description": "Calculate quality scores for memories based on importance, confidence, usage, decay, and connections. Identifies high-quality and needs-attention memories.",
+            "tags": ["curator", "quality", "scoring"],
+            "examples": [
+                "Score memory quality for this project",
+                "Which memories need attention?"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
+        },
+        {
+            "id": "curator_find_orphans",
+            "name": "Curator: Find Orphans",
+            "description": "Find memories with no relationships (orphans). These may need linking or archiving.",
+            "tags": ["curator", "orphans", "cleanup"],
+            "examples": [
+                "Find orphan memories",
+                "Show unconnected memories"
+            ],
+            "inputModes": ["text"],
+            "outputModes": ["text"]
         }
     ]
 }