This enforces a strict discipline around using Memento MCP as your cross-session knowledge graph. The core idea: always search memory before starting work or debugging, use two-dimensional queries (technical topic plus process learnings), and always traverse the full graph with open_nodes and relations, not just the search summaries. It's opinionated about preventing the split-brain problem where you store things in both files and Memento. The value is compounding feedback loops where yesterday's lessons actually inform today's work instead of getting lost between sessions. If you're not already using Memento MCP, this won't help you, it just fails gracefully and tells you to install it.
npx skills add https://github.com/jwilger/agent-skills --skill memory-protocol