Fix validate_animation subpath track parsing (rfind → find) (#371)

* Fix validate_animation subpath track parsing (rfind → find)

validate_animation was extracting the node portion of each track path with
rfind(":"), so a subpath track like "Target:modulate:a" (the shape every
preset_* produces, plus any "position:y" / "modulate:a" / etc. subpath
track) had its property+subpath glued onto the node part. The validator
then looked up "Target:modulate" against root_node, found nothing, and
reported the track as broken even when the target node was perfectly
healthy.

Split on the FIRST colon — the node↔property boundary — instead. The rest
of the string is property+subpath, which validate_animation doesn't need
to inspect (it's only checking that the target node exists).

Add a regression test that creates a Sprite2D sibling of the player,
attaches a "Target:modulate:a" track, and asserts validate_animation
reports valid_count == 1 / broken_count == 0.

Surfaced by Copilot review on #367. Filing here so the cherry-pick stays
clean and the bug fix carries its own test.

https://claude.ai/code/session_011QxzADbf9zHwfZicjzdvCw

* Narrow comment claim: only preset_fade produces subpath tracks

Per Copilot review on #371: of the four motion presets, only preset_fade
emits a subpath track ("Target:modulate:a"). preset_slide / preset_shake /
preset_pulse use plain ":position" or ":scale" paths.

Tighten the comments in animation_values.gd and test_animation.gd so
future readers debugging validator behavior aren't misled.

https://claude.ai/code/session_011QxzADbf9zHwfZicjzdvCw

* Empty commit to retrigger CI (Godot tests / macOS flake)

* Retrigger CI (third attempt; Godot tests / macOS keeps flaking)

* Replace regression test with broken_tracks.node_path shape assertion

Godot's `get_node_or_null` strips the `:property` suffix of a NodePath
natively — so the original `rfind(":")` and the new `find(":")` produce
the same valid/broken classification at runtime. The bug Copilot flagged
on #371 turns out to only be visible in the broken_tracks[].node_path
field for missing-target subpath tracks: rfind surfaces
"MissingTarget:modulate" while find correctly surfaces "MissingTarget".

The previous test asserted valid_count == 1 on a healthy subpath track,
which passed equally with both rfind and find — i.e. it didn't actually
prove the fix. Replace it with a missing-target test that asserts the
broken_tracks[0].node_path field, which is the real behavior change.

The new test also no longer needs a sibling Sprite2D (the missing target
intentionally doesn't exist), so the test is smaller and avoids any
macOS-specific test-runner sensitivity that surfaced on the prior shape.

Comment in animation_values.gd narrowed to match — the fix matters for
diagnostic output, not for the broken/valid classification.

https://claude.ai/code/session_011QxzADbf9zHwfZicjzdvCw

---------

Co-authored-by: Claude <noreply@anthropic.com>

Author: dsarno

plugin/addons/godot_ai/handlers/animation_values.gd

@@ -183,7 +183,14 @@ func validate_animation(params: Dictionary) -> Dictionary:
 
 	for i in anim.get_track_count():
 		var track_path_str := str(anim.track_get_path(i))
-		var colon := track_path_str.rfind(":")
+		# Split on the FIRST colon (node↔property boundary), not the last.
+		# Godot's get_node_or_null strips the ":property" tail natively, so
+		# the valid/broken classification is the same either way — but for
+		# BROKEN tracks the broken_tracks[].node_path field is what callers
+		# read to diagnose the missing node, and rfind would surface
+		# "MissingTarget:modulate" instead of "MissingTarget" for subpath
+		# tracks like the "Target:modulate:a" shape preset_fade emits.
+		var colon := track_path_str.find(":")
 		var node_part: String
 		if colon >= 0:
 			node_part = track_path_str.substr(0, colon)

test_project/tests/test_animation.gd

@@ -1494,6 +1494,39 @@ func test_validate_animation_not_found() -> void:
 	_remove_node(player_path)
 
 
+## Regression: validate_animation must split track paths on the FIRST colon
+## (node↔property boundary), not the last. For broken subpath tracks
+## (target node missing), the broken_tracks[i].node_path field must be the
+## bare node name — not "MissingTarget:modulate" with the property colons
+## preserved — otherwise the diagnostic misleads agents debugging missing
+## targets. Note: Godot's get_node_or_null strips the ":property" tail
+## natively, so the rfind/find difference is only user-visible in this
+## broken_tracks payload, not in the valid/broken classification itself.
+func test_validate_animation_broken_subpath_reports_clean_node_path() -> void:
+	var player_path := _add_player("TestValidateBrokenSubpath")
+	if player_path.is_empty():
+		skip("Scene not ready — _add_player returned empty path")
+		return
+	_handler.create_animation({"player_path": player_path, "name": "fade", "length": 0.5})
+	_handler.add_property_track({
+		"player_path": player_path,
+		"animation_name": "fade",
+		"track_path": "MissingFadeTarget:modulate:a",
+		"keyframes": [
+			{"time": 0.0, "value": 0.0},
+			{"time": 0.5, "value": 1.0},
+		],
+	})
+	var result := _handler.validate_animation({
+		"player_path": player_path, "animation_name": "fade",
+	})
+	assert_has_key(result, "data")
+	assert_eq(result.data.broken_count, 1)
+	assert_eq(result.data.broken_tracks[0].node_path, "MissingFadeTarget",
+		"broken node_path must be the bare node name — not 'MissingFadeTarget:modulate'")
+	_remove_node(player_path)
+
+
 # ─── animation_preset_* — shared helpers ─────────────────────────────────────
 
 ## Add a sibling node to the scene_root. The preset tools resolve target_path