integration/Docs/Blueprint-Text-Export.md

# Blueprint Text Export

Blueprints are stored as binary `.uasset` files that Claude Code cannot read directly. To work around this, a text exporter automatically converts blueprint graphs to readable text files whenever a blueprint is saved in the Unreal editor.

## How It Works

The game module (`FlxIntegrationModuleImpl` in `Source/Integration/Integration.cpp`) hooks into `UPackage::PackageSavedWithContextEvent`. When a blueprint is saved, it iterates each `UEdGraph` in the blueprint and runs `FlxBlueprintExporter` on it. The output is written to `Saved/BlueprintExports/<BlueprintName>/<GraphName>.txt`.

The exporter class (`Source/Integration/BlueprintExporter.h/.cpp`) processes one graph at a time. The constructor runs all passes and the result is available via `GetOutput()`.

## Output Format

The graph file is written to `Saved/BlueprintExports/<BlueprintName>/<GraphName>.txt`. A details file with node-name-to-GUID mappings is written to `Saved/BlueprintExports/<BlueprintName>/DETAILS/<GraphName>.txt`.

Every line in the graph file starts with a keyword, making it easy to parse. The format is:

```
node Event_Tick
    return Output_Delegate, Delta_Seconds
    goto Set_Tick_Delta_Seconds

node Set_Tick_Delta_Seconds
    input Real Tick_Delta_Seconds = Event_Tick.Delta_Seconds
    return Output_Get
    goto CallFunctionByName
```

### Line Keywords

- `node Name` — starts a new node block.
- `input Type Name = Source` — an input data pin. Source is a `Node.Pin` reference, a literal value, `<self>`, or `<default>`.
- `return Pin1, Pin2` — output data pins.
- `goto Target` — exec flow (single output).
- `goto-if PinName Target` — exec flow (multiple outputs, e.g. branch true/false).
- `// comment text` — comment node.

### Special Handling

- String defaults are shown in quotes.
- Variable get nodes are inlined (the variable name appears directly at the point of use rather than as a separate node).
- Knot (reroute) nodes are followed through transparently.

## Node Ordering

Nodes are sorted by a traversal algorithm: find starter nodes (exec output but no exec input), sort them by Y position, then traverse each. The traversal visits a node's inputs first (so data sources appear before their consumers), then emits the node itself, then follows exec outputs. This produces a readable top-down flow. Any unvisited nodes are appended at the end.

## Node Naming

Names are derived from `ENodeTitleType::ListView` titles, sanitized to alphanumeric plus underscores. Duplicates get `_2`, `_3`, etc.