Skip to content

World's first open-source, agentic video production system. 12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files. Turn your AI coding assistant into a full video production studio.

fix(character-reviewer): align success contract with visual_qa pattern

kapil9713905782f704b3ac
diff --git a/tests/contracts/test_character_animation_pipeline.py b/tests/contracts/test_character_animation_pipeline.py
index 9a9f709d..4c765936 100644
--- a/tests/contracts/test_character_animation_pipeline.py
+++ b/tests/contracts/test_character_animation_pipeline.py
@@ -188,11 +188,12 @@ def test_character_animation_smoke_flow(tmp_path):
def test_character_reviewer_success_false_when_qa_finds_issues():
"""
- CharacterAnimationReviewer.success must be False when QA finds issues.
+ CharacterAnimationReviewer surfaces QA failures via status/issues, not success.
- Callers such as the compose-director gate on result.success to decide
- whether to proceed. If success is always True, a broken rig silently
- passes the gate. See PR #166 (closed) and the follow-up fix in #<this PR>.
+ success=True means the tool executed successfully; the QA verdict lives in
+ character_qa_report.status and character_qa_report.issues — matching the
+ pattern used by visual_qa.py (success=True, verdict in validation_passed).
+ compose-director gates on report.status, not result.success.
"""
result = CharacterAnimationReviewer().execute(
{
@@ -205,14 +206,11 @@ def test_character_reviewer_success_false_when_qa_finds_issues():
)
qa_report = result.data["character_qa_report"]
+ assert result.success is True, "tool execution must succeed even when QA finds issues"
assert qa_report["status"] == "revise", (
f"Expected status='revise' for a broken rig, got '{qa_report['status']}'"
)
assert len(qa_report["issues"]) > 0, "Expected at least one issue for a broken rig"
- assert result.success is False, (
- "success must be False when QA finds issues — "
- "callers gate on result.success without inspecting report['status']"
- )
def test_character_style_is_normalized_for_schema(tmp_path):
diff --git a/tools/character/character_animation.py b/tools/character/character_animation.py
index 2842989f..92bd1c7f 100644
--- a/tools/character/character_animation.py
+++ b/tools/character/character_animation.py
@@ -888,12 +888,8 @@ class CharacterAnimationReviewer(BaseTool):
},
}
artifacts = _write_json(inputs.get("output_path"), report)
- # success=False when QA finds issues so callers can gate on result.success
- # without needing to inspect report["status"]. This mirrors the contract
- # asserted in tests/contracts/test_character_animation_pipeline.py and
- # is consistent with how visual_qa.py surfaces validation failures.
return ToolResult(
- success=not issues,
+ success=True,
data={"character_qa_report": report},
artifacts=artifacts,
duration_seconds=round(time.time() - start, 2),