feat: agents build on each other's responses in sequential chain

CrewAI sequential process already passes earlier outputs to later tasks.
Updated task descriptions so agents explicitly reference, agree/disagree,
and build on teammates' input. CAD agent reviews full team discussion
before deciding whether to generate.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agents/crew_orchestrator.py

@@ -169,7 +169,12 @@ class CrewOrchestrator(BaseOrchestrator):
                 f"As the {agent_def.role}, respond to the user's latest message. "
                 f"Keep your response concise (2-4 sentences). "
                 f"Do NOT repeat anything from the conversation history. "
-                f"Add NEW information from your expertise."
+                f"Add NEW information from your expertise.\n\n"
+                f"You can see what previous agents on the team said in this turn. "
+                f"Build on their input — agree, disagree, refine, or add to their "
+                f"points. Reference their suggestions when relevant (e.g. 'Building "
+                f"on the designer\\'s L-bracket suggestion, I\\'d recommend...'). "
+                f"Do NOT repeat what they already covered."
             )
             if agent_id != "cad":
                 task_description += (
@@ -182,9 +187,10 @@ class CrewOrchestrator(BaseOrchestrator):
                 )
             else:
                 task_description += (
-                    "\n\nFirst evaluate if there is enough information to generate "
-                    "a 3D model. You need at minimum: a clear shape/type, key "
-                    "dimensions, and major features. "
+                    "\n\nReview what the other agents have discussed in this turn. "
+                    "Use their suggestions and specs to evaluate whether there is "
+                    "enough information to generate a 3D model. You need at minimum: "
+                    "a clear shape/type, key dimensions, and major features. "
                     "If NOT enough info, start your response with 'NOT READY:' "
                     "followed by a short list of what is still needed. "
                     "If enough info IS available, generate CadQuery Python code. "
@@ -196,7 +202,8 @@ class CrewOrchestrator(BaseOrchestrator):
                 expected_output=(
                     "A concise response from your expert perspective (2-4 sentences)."
                     if agent_id != "cad"
-                    else "Valid CadQuery Python code that assigns result to a cq.Workplane."
+                    else "Valid CadQuery Python code that assigns result to a cq.Workplane, "
+                         "OR a 'NOT READY:' message listing missing information."
                 ),
                 agent=crew_agent,
             )

agents/prompts.py

@@ -71,6 +71,9 @@ def build_orchestrator_system_prompt(
         "- Use the conversation history for context (know what was decided), "
         "but ONLY respond to the user's latest message. Do NOT repeat or "
         "paraphrase anything already said ��� always advance the discussion.\n"
+        "- Agents should build on each other's input — agree, disagree, refine, "
+        "or add to what others said. Reference other agents' suggestions when "
+        "relevant (e.g. 'Based on the designer's L-bracket idea, I'd specify...').\n"
         "- Each agent should add DIFFERENT information. If one agent covers "
         "dimensions, another should cover materials or tooling, not restate dimensions.\n"
         f"- Do NOT include the CAD Coder agent unless the user explicitly uses one of "