claude-memory-plugin

Claude Memory Plugin

A comprehensive memory system for Claude Code - persistent knowledge storage with semantic search, contextual gotcha injection, and graph-based knowledge management.

Overview

The Claude Memory Plugin extends Claude Code with a sophisticated memory system that:

• Stores knowledge persistently across sessions in structured memory files
• Searches semantically using embeddings to find relevant context
• Injects gotchas proactively via hooks when you read files with known pitfalls
• Manages relationships between memories using directed graphs
• Supports multiple scopes (global, project, local, enterprise) for knowledge isolation
• Supports agent-scoped namespaces so specialised agents maintain isolated knowledge bases

Requirements

• Bun >= 1.0.0 (required - used for hooks, CLI, and package management)
• Claude Code >= 2.0.0
• Ollama (optional but recommended, for semantic search with local embeddings)

Quick Start

Installation

Using the Enhance marketplace:

# Add the marketplace (one-time)
/plugin marketplace add GaZmagik/enhance

# Install the plugin
/plugin install claude-memory-plugin

This installs:

• The memory skill for knowledge storage
• Contextual hooks for automatic gotcha injection
• Agent definitions for memory analysis
• Slash commands for memory operations

Basic Usage

# Use the memory CLI (after bun install)
memory help                       # Show available commands
memory list                       # List all memories
memory search "typescript"        # Full-text search
memory semantic "error handling"  # Semantic search by meaning
memory health local               # View memory health
memory stats local                # Graph statistics

v1.1.0 Features

AI-Powered Thinking with --auto and --call

# Auto-select style/agent based on thought content
memory think add "Review security implications" --call claude --auto

# Cross-provider calling (claude, codex, gemini)
memory think add "Analyse performance" --call codex --model gpt-5-codex
memory think counter "Alternative view" --call gemini

# Explicit style and agent
memory think add "Deep dive" --call claude --style Devils-Advocate --agent security-reviewer

Progressive Hints

First 3 invocations of each think command show helpful hints about available flags. Use --non-interactive to suppress.

Cross-Provider Requirements

Ensure the CLI is installed for your chosen provider. If missing, helpful installation instructions are shown:

• claude: Pre-installed with Claude Code
• codex: npm install -g @openai/codex
• gemini: npm install -g @google/gemini-cli

The --auto flag uses a tiered selection strategy: Ollama (if available) → keyword heuristics → default style.

Enhanced Memory Injection

Configure in .claude/memory.local.md to inject decisions and learnings alongside gotchas:

injection:
  types:
    decision:
      enabled: true
      threshold: 0.35
    learning:
      enabled: true
      threshold: 0.4

Hook multipliers adjust thresholds by context (Edit/Write: 0.8×, Bash: 1.2×). See .claude/memory.example.md for full options.

v1.3.0 Features

Agent-Scoped Memories

Agents can now maintain their own isolated memory namespaces, separate from project and global memories:

# Write a memory scoped to an agent
memory write --title "ESM imports require .js extensions" --type learning --agent typescript-expert

# Read agent-scoped memories
memory list --agent typescript-expert
memory search "imports" --agent typescript-expert

# Include shared project memories in results
memory search "imports" --agent typescript-expert --include-shared

# List all known agents
memory agents

Agent memories are stored in .claude/memory/agents/{agent-name}/ (project scope) or ~/.claude/memory/agents/{agent-name}/ (global scope). Each agent gets its own index, graph, and memory files.

Agent CLI Flags

Flag	Description
--agent <name>	Target an agent's memory namespace
--include-shared	Include project/global memories in read operations
--all-agents	Apply to all agents (for listing/stats)
--target-agent <name>	Target agent for cross-agent linking

All existing commands work unchanged without --agent — full backward compatibility.

v1.4.0 Features

Cross-Scope Auto-Linking

The suggest-links command now creates cross-scope links automatically when using --auto-link:

# Suggest and auto-create links within agent scope + shared scopes
memory suggest-links --agent typescript-expert --include-shared --auto-link

# Suggest and auto-create links across ALL scopes (project, global, all agents)
memory suggest-links --all-scopes --auto-link