Memory Graph
by @afgonullu
Agent-agnostic personal knowledge graph stored as markdown files with YAML frontmatter. Use when you need persistent context about the user's life, projects,...
If ~/memory/ does not exist, read references/setup.md and follow the scaffolding steps. This creates the folder structure, symlinks scripts, seeds initial nodes, and builds indexes.
If ~/memory/ already exists, proceed to Operations below.
1. Read before writing. Check if a node exists before creating a duplicate.
2. Log significant work. Append to log/YYYY-MM-DD.md when you do something significant.
3. Use refs in logs. If your work touches existing concepts/projects, add {ref: ...}.
4. Scan for connections. When creating a node, check for related nodes (shared tags, mentions).
5. Never edit indexes/. Run the rebuild script instead.
6. Add fields freely. If a node needs a field that doesn't exist yet, just add it.
7. Add categories freely. Need graph/recipes/? Create the folder. No config changes.
8. Preserve unknown fields. If a node has fields you don't recognize, leave them.
9. Use suggested type for uncertain relations. Let a human or another agent confirm.
10. One-side relations only. Write on whichever node feels natural. Index computes the reverse.
11. Keep the body useful. Frontmatter is for machines. The markdown body is for humans.
12. Set updated on changes. Keep the updated field current for temporal tracking.
13. Use standard relation types. Prefer the taxonomy above. Run lint.js to check.
clawhub install agent-memory-graph