mypy

Author: mcp-release-bot

pyproject.toml

@@ -148,3 +148,9 @@ markers = [
     "slow: marks tests as slow (deselect with '-m \"not slow\"')",
     "integration: marks tests as integration tests",
 ]
+
+[tool.uv]
+dev-dependencies = [
+    "pytest>=8.4.2",
+    "pytest-asyncio>=1.2.0",
+]

src/mcp_as_a_judge/workflow/workflow_guidance.py

@@ -239,28 +239,43 @@ async def calculate_next_stage(
     logger.info(f"Calculating next stage for task {task_metadata.task_id}")
 
     try:
-        # For XS/S tasks, skip planning and provide direct implementation guidance
-        if should_skip_llm_plan_validation(task_metadata):
+        # Check task state first - ALL tasks need user feedback when in CREATED state
+        if task_metadata.state == TaskState.CREATED:
+            return WorkflowGuidance(
+                next_tool="get_user_feedback",
+                reasoning="Task is in CREATED state - user feedback is required to clarify requirements and technical decisions.",
+                preparation_needed=[
+                    "Analyze the repository to detect programming language, frameworks, and existing patterns",
+                    "Analyze current request for requirement gaps and ambiguities",
+                    "Identify technical decisions needed (language, framework, database, API style, etc.)",
+                    "Prepare specific clarifying questions based on repository context",
+                    "Generate suggested options with pros/cons for each decision area",
+                ],
+                guidance=(
+                    f"{_load_todo_guidance()}"
+                    "Call get_user_feedback to gather requirement clarifications and technical decisions. "
+                    "After user feedback, proceed to plan creation and approval workflow."
+                ),
+            )
+
+        # For XS/S tasks that have completed user feedback, skip LLM plan validation
+        if should_skip_llm_plan_validation(task_metadata) and task_metadata.state not in [TaskState.CREATED, TaskState.REQUIREMENTS_FEEDBACK]:
             logger.info(
-                f"Task size {task_metadata.task_size.value} - skipping planning "
+                f"Task size {task_metadata.task_size.value} - skipping LLM plan validation "
                 f"phase, proceeding to implementation"
             )
-            # XS/S tasks skip planning but still need implementation
+            # XS/S tasks skip LLM validation but still need user feedback and approval
             # → code review → testing → completion
-            # For deterministic tests, do not prescribe next tool; provide guidance only
             return WorkflowGuidance(
                 next_tool=None,
                 reasoning=(
                     f"Task size is {task_metadata.task_size.value.upper()} - "
-                    f"planning phase can be skipped for simple fixes and minor "
+                    f"LLM plan validation can be skipped for simple fixes and minor "
                     f"features."
                 ),
                 preparation_needed=[
-                    "Analyze the repository to detect programming language, frameworks, and existing patterns",
-                    "Analyze current request for requirement gaps and ambiguities",
-                    "Identify technical decisions needed (language, framework, database, API style, etc.)",
-                    "Prepare specific clarifying questions based on repository context",
-                    "Generate suggested options with pros/cons for each decision area",
+                    "Implement according to user requirements and feedback",
+                    "Follow established patterns from repository analysis",
                 ],
                 guidance=(
                     f"{_load_todo_guidance()}"