knowledge-ops

Classify the knowledge type first, then store it in the appropriate layer: memory files for quick-access context, MCP memory for semantic search, and a knowledge base Git repo for durable long-form docs. Always search for existing entries before creating new ones to avoid duplicates.

Export sessions from each source and commit them into a central knowledge base repository with a generated session index for browsing. Add short summaries to memory files for quick recall, and redact sensitive data before pushing to Git.

MCP memory is a structured knowledge graph accessed via tools like create_entities, create_relations, add_observations, and search_nodes. Use it for cross-session persistence, semantic search, and mapping relationships between stored facts.

Active execution truth such as roadmaps, rollouts, and implementation state belongs in GitHub issues, PRs, and Linear first. Local notes and the knowledge base repo are for durable synthesized context, not live project state.

Search memory files and query MCP memory with relevant terms before storing anything new, and check GitHub or Linear for existing records. Update the existing entry rather than creating a parallel copy of the same fact.

Memory files suit concise, quick-access context but grow unwieldy for large documents or structured data. Use external stores like Supabase or PostgreSQL for large documents and SQL-queryable data, keeping only summaries in memory.