AnyImage/openspec/changes/optimize-cutout-geometry-nodes/tasks.md

1. Internal attribute naming and wildcard cleanup

• 1.1 Rename CUT_ATTRIBUTE from .o_anyimage_depth_cut to _o_anyimage_depth_cut and update depth_surface.py, image_depth_panorama.py, and asset checks.
• 1.2 Replace the internal split-marker removal path with one wildcard Remove Named Attribute matching _o_* before each depth surface output.
• 1.3 Update tests/tools/nodes/test_temporary_attributes.py, test_depth_surface_split.py, and tools/nodes/check.py to assert the new prefix and remove stale .o_anyimage_* expectations.

2. Reuse the projected outline field

• 2.1 Change build_volume_fields, original_boundary_field, and smooth_cut_boundary to accept an already-evaluated boundary field instead of creating their own edge_boundary_field.
• 2.2 In _build_depth_surface, compute the post-cleanup outline once and pass it to projection-time consumers and boundary smoothing.
• 2.3 Add a node-count or evaluation-equivalence test proving the projected outline consumers share one field on the same topology.

3. Derive center from shared low-frequency fields

• 3.1 Restructure build_volume_fields and build_front_fields so depth_slow and profile_slow are computed once with the existing blur settings.
• 3.2 Replace the direct blur(depth - profile) center computation with depth_slow - profile_slow.
• 3.3 Add a regression test comparing derived center values and Balloon geometry to the previous behavior within floating-point tolerance.

4. Use triangle-only strip cleanup in Depth Cutout

• 4.1 Split _remove_strip_faces or add a triangle-only helper so quad cleanup nodes are built only for non-triangle callers.
• 4.2 Update split_depth_surface callers so O Image Depth Cutout uses the triangle-only path while Depth Plane and Depth Panorama keep the existing quad path.
• 4.3 Update strip cleanup tests to cover both paths and verify Depth Cutout output is unchanged.

5. Short-circuit zero-thickness branches

• 5.1 In image_cutout.py, gate _build_balloon and _build_shell so thickness below the existing zero threshold returns the original surface before extrusion or Repeat.
• 5.2 In image_depth_cutout.py, return the projected front surface when the selected thickness is zero, and only build thicken_depth_surface plus shell field sampling for positive thickness.
• 5.3 Add tests verifying zero-thickness modes produce the single surface, preserve UV/public attributes, and emit no thickness-related node warnings.

6. Rebuild assets and run regression

• 6.1 Run the focused Cutout and Depth Cutout test files and fix any behavioral differences.
• 6.2 Run uv run node-group build to rebuild and validate O_AnyImage.blend.
• 6.3 Run the full uv run pytest suite and confirm normal exit with no new expected failures.
• 6.4 Review git diff --check and the final node/attribute inventory to confirm no stale .o_anyimage_* references remain.