Overview
Memory Search Skill is the memory-search skill from AI Maestro, an MIT-licensed dashboard by 23blocks for orchestrating many coding agents (Claude Code, Codex, and others) with persistent memory, agent-to-agent messaging, and multi-machine support. The skill is published in the ai-maestro-plugins repository and requires a running AI Maestro installation with Bash access.
Its instructions are deliberately forceful: at the start of any non-trivial task, the agent must search memory for previous discussions, past decisions, prior context, and earlier work before doing anything else, and check memory before re-reading files to re-learn something. Files show the current state; memory records why it is that way.
How It Works
AI Maestro classifies past sessions into long-term memory cards (decisions, facts, preferences, patterns, insights) and keeps the raw conversation history. It also injects a ## Memory: block at session start and with each prompt. memory-search.sh returns long-term memories first, then matching raw history. Entities mentioned in memories are graph nodes, so you can ask for everything known about one host, service, file, or person.
| Command | Purpose |
|---|---|
memory-search.sh "<query>" | Hybrid search (default) |
memory-search.sh --about "<entity>" | Relations and memory cards for one entity |
--mode semantic | Conceptually related results |
--mode term | Exact term matching |
--mode symbol | Code identifiers mentioned in past conversations |
--role user / --role assistant | Only your messages or only the agent's |
Quick Start
curl -fsSL https://raw.githubusercontent.com/23blocks-OS/ai-maestro/main/scripts/remote-install.sh | sh
The installer sets up the dashboard (at http://localhost:23000), the messaging system, and the Claude Code plugin with its skills and CLI scripts. It needs Node.js 18+ and tmux; Windows is supported through WSL2.
Use Cases
- "Continue working on X" across sessions without re-explaining context.
- Respecting decisions made weeks ago by another session or agent.
- Finding where a function was discussed with
--mode symbol.
Limitations
- Only works inside AI Maestro; it is not a standalone memory tool.
- Memories are excerpts from earlier sessions and can be outdated, so the skill tells agents to verify before acting.
FAQ
Is the memory search automatic?
The skill instructs the agent to search proactively on every task, and AI Maestro also injects relevant memory into the prompt.
Alternatives
- agentmemory, local persistent memory for coding agents via MCP and hooks.
- Letta, a platform for stateful agents with long-term memory.
Related
Related AI Maestro skills: Docs Search, Graph Query, and Agent Messaging. More in Skills.
Comments
No comments yet. Be the first to comment!
Related Tools
Related Insights
Claude Code's Next Frontier: Not Code, But Your Local Obsidian Knowledge Base
Explore how Obsidian + Claude Code transforms from a knowledge management tool into your private AI assistant. Complete guide including obsidian-skills, Claudian plugin, Claudesidian template, and best practices for achieving both data privacy and AI capabilities.
Anthropic Subagent: The Multi-Agent Architecture Revolution
Deep dive into Anthropic multi-agent architecture design. Learn how Subagents break through context window limitations, achieve 90% performance improvements, and real-world applications in Claude Code.
Skills + Hooks + Plugins: How Anthropic Redefined AI Coding Tool Extensibility
An in-depth analysis of Claude Code's trinity architecture of Skills, Hooks, and Plugins. Explore why this design is more advanced than GitHub Copilot and Cursor, and how it redefines AI coding tool extensibility through open standards.