Version 2026.7.1

[Mustard2]

• Enhancement: Some Creator Tools now offer the possibility to remove unused data (null shape keys, unused vertex groups, attributes, UV Maps).
• Enhancement How to setup the new Jiggle physics? #399: The Jiggle Cage (Quick) operator now infers the direction automatically by default.
• Fix Custom Properties bug #402: Improved the section selection for custom properties in Configuration mode, in order to be less confusing and restore the custom property list correct icons indentation.
• Fix Create Morphs Preset error #403: Morphs with shape keys might fail to be written.
• Fix: Context in active model selection.

[Mustard2]

@claude Review

[claude]

Claude finished @Mustard2's task in 3m 48s —— View job

---

Code Review — PR #404 (Version 2026.7.1)

Reviewed the diff against origin/main (13 files, +446/-80). Overall the changes are clean and match the repo conventions in CLAUDE.md (register/unregister ordering, mustardui_active_object usage, ops_*/settings_* naming). Two issues found, both minor:

1. Dead fallback / wrong comparison in the custom-property section icon (custom_properties/ui_list.py:44-47)

section = rig_settings.body_custom_properties_sections.get(item.section)
icon = "RECORD_OFF"
if section is not None:
    icon = section.icon if section not in {"", "NONE"} else "DOT"

section is a MustardUI_SectionItem PropertyGroup instance, not a string, so section not in {"", "NONE"} is always True — the "DOT" fallback branch is unreachable, and a section with an unset icon (""/"NONE") won't fall back to "DOT" as intended. The correct comparison (and the one already used a few lines away in the new custom_properties/ops_set_section.py:23) is against section.icon, not section itself:

icon = section.icon if section.icon not in {"", "NONE"} else "DOT"

Since this PR's own goal (#402) is to "restore the custom property list correct icon indentation," this is worth fixing for consistency even though it isn't a crash (icon values are constrained by the EnumProperty, so nothing breaks — it's a cosmetic/logic inconsistency).

Fix this →

2. Missing model_body null guard in the new shape-key preset paths (morphs/settings_presets.py:20,65, morphs/ops_defvalue.py:40)

The #403 fix correctly switches from the (non-existent) rig_settings.data to rig_settings.model_body.data, but none of the three call sites check rig_settings.model_body is not None before dereferencing .data:

shape_keys = rig_settings.model_body.data.shape_keys  # settings_presets.py:20, ops_defvalue.py
...
elif m.get("shape_key") and rig_settings.model_body.data.shape_keys:  # settings_presets.py:65

model_body is a PointerProperty that can legitimately be None — the codebase guards against this explicitly elsewhere (configuration/ops_quick_setup.py, configuration/ops_configuration.py, misc/get_ui_objects.py, etc.). In practice these preset/morph paths are only reachable on an enabled/configured model where model_body should already be set, so this is likely low-risk, but an unguarded AttributeError would be a regression vs. the previous defensive rig_settings.data and rig_settings.data.shape_keys check pattern that was dropped.

Fix this →

Other notes (no action needed)

• model_selection/active_object.py: switching bpy.context.scene → context.scene is a correct fix — using the passed-in context instead of the global is the right call when this helper can be invoked from contexts where they diverge (matches the "Fix: Context in active model selection" changelog entry).
• tools_creators/ops_jiggle.py: the automatic Pin-direction inference (border-vertex vs. island-center vector, then nearest-island matching for the generated cages) looks correct, and the refactored gradient projection math is behavior-preserving for the existing manual +X/-X/… options (verified the sign/inversion algebra matches the old per-axis branches).
• misc/mesh_cleanup.py, ops_optimize_sk.py, ops_collision_cage.py, ops_jiggle_accurate.py: the new cleanup helpers (void shape-key detection/removal, unused vertex groups/attributes) are well-isolated, skip morph-managed shape keys correctly, and are wired behind opt-in operator toggles.

Not run: ruff check . / ruff format --check . — Bash execution requires interactive approval in this environment and none was available; please run these locally/in CI to confirm lint/format still pass (the CI lint job will also catch this).

[Mustard2]

@claude Review

[claude]

Claude finished @Mustard2's task in 3m 12s —— View job

---

Code Review — PR #404 (Version 2026.7.1)

• Gather context (PR diff, previous review comments)
• Review changed files against origin/main
• Check whether previously-flagged issues were addressed
• Post final review

I re-reviewed the full diff (git diff origin/main...HEAD, 23 files, +677/-108), including the 5 commits pushed after the previous automated review (10f003a…08054bf).

Previously-flagged issues — status

1. Icon fallback in custom_properties/ui_list.py — ✅ fixed. Line 47 now correctly compares section.icon, not section itself.
2. Missing model_body null guard in morphs/settings_presets.py / morphs/ops_defvalue.py — still unaddressed, but confirmed low-risk: configuration/ops_configuration.py:37 requires rig_settings.model_body to be set before a model can be enabled, and all these code paths gate on config=0 (enabled model only), so model_body should never be None when reached in practice.

New finding

tools_creators/ops_optimize_sk.py:18 — add_shape_key_mute_driver default flip breaks the "revert" (remove drivers) button on first use

This PR (commit b8bb0ab) changed the property default from True to False:

add_shape_key_mute_driver: bpy.props.BoolProperty(
    default=False,   # was True
    ...
)

This same property is reused for both directions of the operator: the forward "Optimize" call and the "revert" call (menu/menu_tools_creators.py:271-273, .revert = True). In revert mode, the dialog draws only this one checkbox (ops_optimize_sk.py:164-169), and execution is gated by:

if not self.add_shape_key_mute_driver and not remove_void:  # remove_void is always False when reverting
    self.report({"WARNING"}, "MustardUI - No Option Selected.")
    return {"CANCELLED"}

Since Blender operators reuse the last-set property values across invocations of the same bl_idname, the very first time a user clicks the revert (LOOP_BACK) button in a session, add_shape_key_mute_driver is still at its unset default (False), so confirming the dialog immediately cancels with "No Option Selected" instead of removing the mute drivers — the user has to notice this, re-open the dialog, and manually check the box. Previously (default True) this worked out of the box. Low severity (recoverable, not a crash), but likely an unintended side effect of reusing this property for the new opt-in "Remove Void Shape Keys" logic.

Suggested fix: give the revert-mode checkbox its own property (or force add_shape_key_mute_driver = True when self.revert is set in invoke()), so the "remove drivers" default isn't affected by the new forward-flow default.

Fix this →

Other notes (no action needed)

• custom_properties/ops_set_section.py, ops_menu_settings.py: the new per-property section picker is wired correctly — sections_enum matches the item-caching pattern needed to avoid EnumProperty string GC, index passed to MustardUI_Property_SetSection matches the true arm.MustardUI_CustomProperties collection index (verified against menu/menu_configure_body.py's template_list binding), and custom_prop.section is a plain StringProperty so the ""/SECTION_NONE round-trip is safe.
• model_selection/active_object.py: bpy.context.scene → context.scene is correct.
• physics/settings_presets.py: removal of the unused force parameter from apply_physics_preset is clean — no remaining call sites pass it.
• presets/ops_create.py, ops_delete.py, ops_transfer.py: adding options={"HIDDEN"} to preset_type is a reasonable cleanup, no behavior change.
• misc/mesh_cleanup.py, tools_creators/ops_collision_cage.py, tools_creators/ops_jiggle_accurate.py: the new opt-in cleanup helpers (void shape-key/vertex-group/attribute removal) are well-isolated, correctly preserve modifier-referenced vertex groups (e.g. Pin groups) via the keep= parameter, and are gated behind their own operator toggles.
• tools_creators/ops_transfer_vertex_groups.py: new "All Groups" / "From Bones" bulk-add helpers are straightforward and correctly dedupe against the existing list.

Not run: ruff check . / ruff format --check . — Bash execution requires interactive approval in this environment; please confirm CI's lint job passes.

[Mustard2]

@claude Review now again

[claude]

Claude finished @Mustard2's task in 1m 49s —— View job

---

Code Review — PR #404 (Version 2026.7.1)

• Gather context (latest PR diff, previous review comments)
• Check whether previously-flagged ops_optimize_sk.py revert-default issue was addressed
• Review the 3 new commits since the last review (1828901, e8bdb62, b01d33c)
• Post final review

Re-reviewed git diff origin/main...HEAD (23 files, +749/-108). Since the last automated review, 3 new commits landed:

Previously-flagged issue — status

tools_creators/ops_optimize_sk.py revert-default flip — ✅ fixed in 1828901. invoke() now forces self.add_shape_key_mute_driver = True when self.revert is set, so the revert (remove drivers) button works correctly on first use again, independent of the new forward-flow default (False).

New commits reviewed

• e8bdb62 (comments cleanup) — pure docstring/comment trimming in misc/mesh_cleanup.py, no behavior change.
• b01d33c (broadened scope of vertex group mesh cleanup) — adds collect_used_vertex_groups(), replacing the old "only check modifier.vertex_group" logic with a generic recursive walk (_vertex_group_properties, depth-limited to 3, skipping ID datablocks and large data collections via MAX_WALKED_COLLECTION) plus dedicated handling for Geometry Nodes modifier inputs (modifier.values()), particle systems, soft body settings, shape-key vertex groups, and constraint subtargets (including bone constraints on any object in the file). This is a correctness improvement for clear_unused_vertex_groups — it now avoids deleting vertex groups that are legitimately referenced from more places than before. The implementation is conservative by design (a same-named string custom property on a modifier could theoretically cause a vertex group to be over-preserved via the modifier.values() scan), but since this only makes cleanup more cautious rather than less, it's a safe direction for an opt-in destructive tool. No bugs found in the recursion bounds, None-guards (obj.soft_body, other.pose), or the single call site (tools_creators/ops_jiggle_accurate.py:1113-1116, which still correctly passes keep=[pin_name, structural_group_name] on top of the new automatic detection).

No new issues found. Nothing else outstanding from prior review rounds beyond the already-noted, confirmed-low-risk model_body null-guard observation (not a blocker — all reachable call paths are gated on an enabled/configured model).

Not run: ruff check . / ruff format --check . — Bash execution requires interactive approval in this environment; please confirm CI's lint job passes before merging.
· branch 2026.7.1