What is the Memory Protocol skill for Claude Code?

It is a specialized capability that allows Claude to systematically store and recall project-specific knowledge across different chat sessions using a Memento MCP knowledge graph.

Does this skill require external tools?

Yes, it requires the Memento MCP server to be installed and configured. Without these tools, the skill cannot persist memory and will default to standard behavior.

How does multi-dimensional search improve performance?

It forces the agent to search for both technical solutions (how to fix a bug) and process learnings (how to avoid workflow mistakes), preventing the repetition of previous procedural errors.

Can this skill handle agent crashes?

Yes, it utilizes a WORKING_STATE.md file to track progress during long-running sessions, allowing the agent to recover context after a crash or compaction event.

Does it replace project documentation?

No, it is intended to complement documentation by capturing 'mid-term' memory like debugging wins, specific preferences, and temporary workarounds that don't belong in formal docs.

Cross-Session Memory Protocol

Name: Cross-Session Memory Protocol
Author: jwilger

byjwilger

•

生産性とワークフロー

Implements persistent knowledge storage and retrieval using Memento MCP to maintain project context across multiple AI agent sessions.

The Memory Protocol skill solves the problem of context loss between AI coding sessions by leveraging the Memento MCP knowledge graph. It establishes a rigorous workflow where the agent searches for both technical solutions and process-related learnings before acting, ensuring that past debugging efforts, architectural decisions, and user preferences are never forgotten. By transforming transient chat history into a structured, traversable graph, it enables compound feedback loops that accelerate development and prevent the repetition of previous mistakes through a systematic three-step recall process.

主な機能

01Crash recovery infrastructure using WORKING_STATE.md tracking

022 GitHub stars

03Automatic storage of solutions, conventions, and user corrections

04Persistent cross-session knowledge storage via Memento MCP knowledge graph

05Mandatory multi-dimensional search covering technical and workflow domains

06Three-step recall strategy (Search, Open, and Traverse) for deep context

ユースケース

01Maintaining complex architectural context across long-running development projects

02Onboarding subagents to project-specific conventions and user preferences

03Preventing repeated debugging of environment-specific issues or tool quirks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jwilger/agent-skills memory-protocol

For use in Claude.ai and ChatGPT

Download Skill