knowledge-ops

To synchronize knowledge across local files, Claude memory, and vector stores, this Skill classifies content and routes it to the appropriate layer. It then updates existing records, refreshes indexes, and commits summarized artifacts to a knowledge base repository.

Deduplicating knowledge from conversation sessions and project memory requires searching existing memory and MCP nodes before creating new entries. This workflow updates existing records instead of duplicating data, keeping your knowledge base clean and queryable.

Knowledge ingestion with an MCP memory graph and Git repositories works by classifying content based on durability needs, creating MCP entities and relations, redacting sensitive data, and committing synthesized session exports or long-form notes to the KB repo.

Yes, you can redact sensitive data before committing session exports to a knowledge base. The sync and commit workflows specifically mirror active execution truth to GitHub or Linear, apply redaction, and then commit synthesized notes to the repository.

This approach fully supports multi-layer ingestion across local files, vector stores, and external databases. It classifies and routes content into the appropriate layer—quick memory files, MCP graph, KB repo, or external DB—based on specific type and durability needs.

You should avoid multi-layer knowledge management workflows for simple, ephemeral tasks that do not require durable context or cross-machine recall. It is designed for complex project workflows needing to capture conversations, documents, and workspace state across synchronized stores.