Knowledge Graph Memory Education Edition

Problem

When memory is stored as flat markdown notes, it is hard to model statements such as X belongs to Y, reverse lookup from a knowledge point to poems, or reusable cross-session aggregation. This skill treats state as a typed graph of entities, properties, and relations, with replayable event logs.

How It Works

• Entity-based memory: objects have type, properties, and relations, such as Poem, KnowledgePoint, and Student.
• Append-only storage: writes go to memory/ontology/graph.jsonl; final graph state is derived by replaying events, supporting backup, restore, and repair.
• Query and validation: supports property reverse lookup, relation traversal, and constraint checks; schema.yaml can define required fields, enums, and relation direction.
• Bulk import: for large datasets, generate jsonl event lines directly instead of spawning a CLI for each item; examples cover poem and classical-text imports.

Boundaries

Best for structured state where reverse lookup, aggregation, and relation reuse matter. It is not a substitute for scratch notes, simple config, or strongly consistent production databases. Generated id values should not be hardcoded; high-frequency concurrent writes should use a single writer or migrate to SQLite.