---

name: core-graph-engine description: "Guides Meshroom graph, node, attribute, serialization, compatibility, cache, execution, and status workflows for agents working with .mg project graphs or Meshroom core APIs." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Meshroom Core Graph Engine

Use this sub-skill when a task is about Meshroom's in-memory graph model, .mg serialization, node/attribute instances, DAG traversal, compatibility nodes, invalidation, cache directories, or local graph execution.

Read First

• API reference for verified Graph, Node, Attribute, graph IO, status, and compatibility objects.
• Graph workflows for common recipes: create/connect/save/load, import graph content, configure outputs, execute a safe graph, and inspect status.
• Troubleshooting for compatibility nodes, stale cache, invalid edges, submitted/running conflicts, and chunk failures.

Route by Task

Task	Use
Add nodes, connect attributes, rename/find nodes, compute topological order	Graph workflows
Save/load .mg files, save templates, import partial graph content	Graph workflows
Explain .mg JSON headers, node version metadata, minimal template serialization	API reference
Diagnose CompatibilityNode, UnknownNodeType, VersionConflict, DescriptionConflict	Troubleshooting
Understand node status, chunks, preprocess/process/postprocess, cache folders	API reference and Troubleshooting
Execute a graph locally or reason about dfsToProcess	Graph workflows

Core Mental Model

• A Graph is a DAG of node instances connected by edges from output attributes to input attributes.
• Node descriptors live in meshroom.core.desc; node instances and valued attributes live in meshroom.core.node and meshroom.core.attribute.
• A connected input reads from its upstream output. Attribute values participate in UIDs and invalidation unless the descriptor disables invalidation.
• Computation runs by nodes and chunks. executeGraph selects nodes that are not already successful, initializes statuses, runs preprocess/chunks/postprocess, and saves the graph before compute.
• Graph files are JSON .mg files with a header and graph payload. Template saves omit default inputs and runtime output/cache details.

Safe API Skeleton

import meshroom
meshroom.setupEnvironment()
import meshroom.core
from meshroom.core.graph import Graph, loadGraph

meshroom.core.initNodes()
graph = Graph("example")
node = graph.addNewNode("InputString", name="Text_1")
node.string.value = "hello"
graph.save("example.mg")
loaded = loadGraph("example.mg", strictCompatibility=False)
print([n.name for n in loaded.nodes])

If you are creating custom node types for this skeleton, route to node-descriptors first.

Decision Points

• Use loadGraph(path, strictCompatibility=True) only when compatibility nodes should be a hard failure.
• Use Graph.configureOutputNodes([...]) only after a graph/template includes OutputNode instances.
• Use dfsOnFinish/dfsOnDiscover for topology inspection; use dfsToProcess for compute planning because it skips already successful branches.
• Treat forceStatus and forceCompute as deliberate overrides. They can clobber useful status/cache evidence.

Verification Anchors

Native tests that exercise this route include graph ordering, graph IO, compatibility/upgrade, compute status, invalidation, and attribute callback cases. Prefer focused tests over full pipelines:

pytest tests/test_graph.py -q
pytest tests/test_graphIO.py -q
pytest tests/test_compatibility.py -q
pytest tests/test_compute.py -q

Run them from a Meshroom checkout only when doing repository maintenance. For package usage questions, use small synthetic graph snippets instead of requiring the original checkout.