fix(skills): correct four API claims that make agent code silently wrong

rules/use-foreach-set-for-bulk-data.mdc

@@ -66,8 +66,12 @@ Same 100,000 vertices: under a millisecond.
 
 The signature is `foreach_set(attribute, sequence)` where:
 
-- `attribute` is the name of the per-element attribute as a string
-  (`"co"`, `"normal"`, `"index"`, `"use_smooth"`, etc.).
+- `attribute` is the name of a **writable** per-element attribute as a
+  string (`"co"`, `"use_smooth"`, `"material_index"`, etc.). Read-only
+  attributes such as vertex `"normal"` and `"index"` accept `foreach_set`
+  without error and change nothing (measured on 5.2.1 LTS): vertex normals
+  are derived, so read them with `foreach_get` (or `mesh.vertex_normals`)
+  and write custom normals with `mesh.normals_split_custom_set(...)`.
 - `sequence` is a flat 1D iterable of `len(elements) * components`
   values. For vector attributes (3 floats per vertex), the buffer is
   3x as long as the element count.
@@ -85,8 +89,6 @@ import numpy as np
 
 n = len(mesh.vertices)
 coords = np.empty(n * 3, dtype=np.float32)
-mesh.vertices.foreach_get("co", coords)
-
 mesh.vertices.foreach_get("co", coords)
 xs = coords[0::3]
 ys = coords[1::3]

skills/drivers-and-app-handlers/SKILL.md

@@ -210,21 +210,24 @@ from bpy.app.handlers import persistent
 
 @persistent
 def increment_save_count(filepath):
-    # save_post passes the saved file path (a string) as its only argument.
-    # Store the running count on the current scene.
+    # save_pre runs before the write, so the new count lands in the file being
+    # saved (a save_post write would only reach the next save). It receives the
+    # target path. ID-property keys are limited to 63 characters, so key by the
+    # file name, never the full path, which raises KeyError when longer.
     scene = bpy.context.scene
+    key = bpy.path.basename(filepath)[:63] or "untitled"
     counts = scene.get('save_counts', {})
-    counts[filepath] = counts.get(filepath, 0) + 1
+    counts[key] = counts.get(key, 0) + 1
     scene['save_counts'] = counts
 
 
 def register():
-    bpy.app.handlers.save_post.append(increment_save_count)
+    bpy.app.handlers.save_pre.append(increment_save_count)
 
 
 def unregister():
-    if increment_save_count in bpy.app.handlers.save_post:
-        bpy.app.handlers.save_post.remove(increment_save_count)
+    if increment_save_count in bpy.app.handlers.save_pre:
+        bpy.app.handlers.save_pre.remove(increment_save_count)
 ```
 
 ### Worked example: cleanup on exit (5.1+)

skills/engine-export-presets/SKILL.md

@@ -51,6 +51,7 @@ def apply_selected_mesh_transforms():
 ```python
 bpy.ops.export_scene.gltf(
     filepath=path,
+    export_format="GLB",
     use_selection=True,
     export_yup=True,
     export_apply=True,
@@ -68,6 +69,7 @@ Snippet: [`snippets/export_preset_unity.py`](https://github.com/TMHSDigital/Blen
 ```python
 bpy.ops.export_scene.gltf(
     filepath=path,
+    export_format="GLB",
     use_selection=True,
     export_yup=False,
     export_apply=True,
@@ -115,7 +117,7 @@ Snippet: [`snippets/export_preset_unreal.py`](https://github.com/TMHSDigital/Ble
 
 Probed on 4.5 LTS, 5.1, and 5.2: `export_yup` on glTF and `axis_forward` / `axis_up` / `global_scale` on FBX are present on all three. No version branch for the axis kwargs. Guard by requiring those names in operator RNA so a future rename fails loudly, as [`examples/gltf-export-roundtrip/`](https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/examples/gltf-export-roundtrip) does.
 
-`export_format` defaults differ by call site; pass the filepath suffix (`.glb` / `.gltf` / `.fbx`) and let the operator infer, or set `export_format` explicitly on glTF.
+Always pass `export_format` on glTF and make the filepath suffix match it. The operator does **not** infer the format from the suffix: `export_scene.gltf(filepath="x.gltf")` with no `export_format` writes a binary `x.glb`, so the requested path never exists (measured on 4.5.11 and 5.2.1). Use `export_format="GLB"` with `.glb`, or `"GLTF_SEPARATE"` with `.gltf`.
 
 ## Related
 

skills/geometry-nodes-python/SKILL.md

@@ -164,8 +164,10 @@ To list all available Geometry node types in your Blender version:
 
 ```python
 import bpy
-for cls in bpy.types.GeometryNode.__subclasses__():
-    print(cls.bl_idname)
+# bpy.types registers node classes lazily, so GeometryNode.__subclasses__()
+# returns [] until something touches them. dir(bpy.types) lists them all.
+for name in sorted(n for n in dir(bpy.types) if n.startswith("GeometryNode")):
+    print(name)
 ```
 
 ## Wiring nodes
@@ -323,7 +325,7 @@ def has_for_each_element():
    tree.nodes.new('GeometryNodeSetPosition')  # RIGHT
    ```
 
-   When in doubt, list `bpy.types.GeometryNode.__subclasses__()` and grep.
+   When in doubt, list `[n for n in dir(bpy.types) if n.startswith("GeometryNode")]` and grep. (`bpy.types.GeometryNode.__subclasses__()` returns an empty list: the classes are created lazily.)
 
 3. **Linking nodes by name when both inputs share a name** (`Value`, `Vector`, `Geometry`) and getting the wrong one. Use indices or the named inputs of the parent node.
 

snippets/export_preset_godot.py

@@ -31,6 +31,7 @@ def export_preset_godot(filepath, selected_only=True, draco=False):
     apply_selected_mesh_transforms()
     bpy.ops.export_scene.gltf(
         filepath=filepath,
+        export_format="GLB",
         use_selection=selected_only,
         export_yup=False,
         export_apply=True,

snippets/export_preset_unity.py

@@ -31,6 +31,7 @@ def export_preset_unity(filepath, selected_only=True, draco=False):
     apply_selected_mesh_transforms()
     bpy.ops.export_scene.gltf(
         filepath=filepath,
+        export_format="GLB",
         use_selection=selected_only,
         export_yup=True,
         export_apply=True,

snippets/export_preset_unreal.py

@@ -41,6 +41,7 @@ def export_preset_unreal_gltf(filepath, selected_only=True, draco=False):
     apply_selected_mesh_transforms()
     bpy.ops.export_scene.gltf(
         filepath=filepath,
+        export_format="GLB",
         use_selection=selected_only,
         export_yup=True,
         export_apply=True,

snippets/gltf_draco_export.py

@@ -14,6 +14,7 @@
 def export_gltf_draco(filepath, selected_only=True, yup=True):
     bpy.ops.export_scene.gltf(
         filepath=filepath,
+        export_format="GLB",
         use_selection=selected_only,
         export_draco_mesh_compression_enable=True,
         export_draco_mesh_compression_level=6,

templates/ai-asset-pipeline-template/pipeline.py

@@ -260,6 +260,7 @@ def export_preset(filepath, preset, draco):
     if preset == "unity":
         bpy.ops.export_scene.gltf(
             filepath=filepath,
+            export_format="GLB",
             use_selection=True,
             export_yup=True,
             export_apply=True,
@@ -270,6 +271,7 @@ def export_preset(filepath, preset, draco):
     if preset == "godot":
         bpy.ops.export_scene.gltf(
             filepath=filepath,
+            export_format="GLB",
             use_selection=True,
             export_yup=False,
             export_apply=True,
@@ -296,6 +298,7 @@ def export_preset(filepath, preset, draco):
     apply_selected_mesh_transforms()
     bpy.ops.export_scene.gltf(
         filepath=filepath,
+        export_format="GLB",
         use_selection=True,
         export_yup=True,
         export_apply=True,