| name | understand-memory-sync |
| description | Sync knowledge graph entities to Memory MCP for cross-session codebase awareness |
| argument-hint | ["project-path"] |
/understand-memory-sync
Sync the most important entities from a project's knowledge graph to the Memory MCP server, enabling cross-session codebase recall without re-analysis.
Instructions
-
Locate the knowledge graph:
- If
$ARGUMENTS contains a path, use that as the project directory
- Otherwise, use the current working directory
- Read
.understand-anything/knowledge-graph.json
- If not found, tell the user to run
/understand first
-
Extract key entities (top ~50-100 nodes by architectural importance):
- All
module nodes (architectural boundaries)
- All
class nodes tagged as agent, core-module, or governance
- File nodes with >200 lines (complexity hotspots)
- Nodes referenced in the tour steps
- The project node itself
-
Create Memory MCP entities using mcp__memory__create_entities:
- Entity name:
{project}:{node.name} (e.g., CoreMind:CoreMind)
- Entity type: map node type → Memory entity type (
module → Architecture, class → Component, file → File, function → Function)
- Observations: node summary, tags, layer assignment, line count, file path
-
Create Memory MCP relations using mcp__memory__create_relations:
- For each
imports edge between synced nodes → imports relation
- For each
inherits edge → extends relation
- For each
depends_on edge between modules → depends_on relation
- For each
contains edge (module → key class) → contains relation
-
Add a project-level observation via mcp__memory__add_observations:
- Entity:
CoreMind
- Observations:
- "Knowledge graph: {N} nodes, {M} edges, {L} layers"
- "Analyzed at: {timestamp}, commit: {hash}"
- "Dashboard: cd ~/Projects/understand-anything/understand-anything-plugin/packages/dashboard && GRAPH_DIR=/path/to/project npx vite --open"
-
Report what was synced:
Synced to Memory MCP:
- {N} entities created
- {M} relations created
- Key modules: {list}
- Key components: {list}
These entities are now available across sessions via mcp__memory__search_nodes.
When to Re-Sync
Re-sync after running /understand on new changes (incremental update). The skill should check if entities already exist and update observations rather than creating duplicates.