modified Dockerfile, inference.py, app/env.py; included openenv.yaml, server7860.py

Dockerfile

@@ -10,4 +10,4 @@ COPY . .
 
 RUN pip install --no-cache-dir -r requirements.txt
 
-CMD ["python", "inference.py"]
+CMD ["sh", "-c", "if [ \"$RUN_MODE\" = \"server\" ]; then uvicorn server:app --host 0.0.0.0 --port 7860; else python inference.py; fi"]

app/env.py

@@ -79,10 +79,10 @@ class CustomerSupportEnv:
 
         reward = 0.0
         done = False
-        #info = {}
-        info = {
-        "final_score": self._compute_final_score() if done else None
-        }
+        info = {}
+        #info = {
+        #"final_score": self._compute_final_score() if done else None
+        #}
 
         collected = self.state_data["collected_info"]
         required = self.state_data["required_info"]

inference.py

@@ -6,12 +6,28 @@ from agent_llm import get_action
 from app.env import CustomerSupportEnv
 
 
+#"""
+#def format_action(action: dict) -> str:
+    #"""Convert action dict → string"""
+#    if not action:
+#        return "null"
+#    return str(action).replace("\n", "").replace("  ", " ")
+#"""
+
 def format_action(action: dict) -> str:
-    """Convert action dict → string"""
     if not action:
         return "null"
-    return str(action).replace("\n", "").replace("  ", " ")
 
+    action_type = action.get("type")
+
+    if action_type == "ask_info":
+        return f"ask_info('{action.get('field')}')"
+    elif action_type == "resolve":
+        return "resolve()"
+    elif action_type == "classify":
+        return "classify()"
+
+    return str(action)
 
 def main():
 
@@ -83,12 +99,19 @@ def main():
         # =========================
         rewards_str = ",".join(f"{r:.2f}" for r in rewards)
 
+        score = 1.0 if success else 0.0
+
+        #print(
+        #    f"[END] success={'true' if success else 'false'} "
+        #    f"steps={step_count} "
+        #    f"rewards={rewards_str}"
+        #)
         print(
-            f"[END] success={'true' if success else 'false'} "
-            f"steps={step_count} "
-            f"rewards={rewards_str}"
+        f"[END] success={'true' if success else 'false'} "
+        f"steps={step_count} "
+        f"score={score:.2f} "
+        f"rewards={rewards_str}"
         )
 
-
 if __name__ == "__main__":
     main()

openenv.yaml

@@ -0,0 +1,116 @@
+name: customer-support-agent
+description: >
+  A goal-oriented customer support environment where an agent must gather
+  required information from the user and resolve the ticket efficiently.
+
+version: 1.0
+
+# =========================
+# ENTRYPOINT
+# =========================
+entry_point: server7860:app
+
+# =========================
+# INTERFACE
+# =========================
+interfaces:
+  - type: http
+    port: 7860
+    routes:
+      reset: /reset
+      step: /step
+
+# =========================
+# ENVIRONMENT LIMITS
+# =========================
+max_steps: 10
+
+reward_range:
+  min: -1.0
+  max: 2.0
+
+# =========================
+# ACTION SPACE
+# =========================
+actions:
+  - name: ask_info
+    description: Request missing information from the user
+    parameters:
+      field:
+        type: string
+        enum:
+          - order_id
+          - account_email
+          - device_type
+          - browser
+
+  - name: classify
+    description: Classify the user issue into a category
+    parameters: {}
+
+  - name: resolve
+    description: Resolve the ticket if sufficient information is available
+    parameters: {}
+
+# =========================
+# OBSERVATION SPACE
+# =========================
+observation_space:
+  type: object
+  properties:
+    ticket_id:
+      type: string
+
+    customer_message:
+      type: string
+
+    history:
+      type: array
+      items:
+        type: object
+
+    known_info:
+      type: object
+      additionalProperties: true
+
+    required:
+      type: array
+      items:
+        type: string
+
+    missing_required:
+      type: array
+      items:
+        type: string
+
+    info_progress:
+      type: number
+
+    status:
+      type: string
+      enum: [open, resolved]
+
+    step_count:
+      type: integer
+
+    remaining_steps:
+      type: integer
+
+# =========================
+# TERMINATION CONDITIONS
+# =========================
+termination:
+  success:
+    description: Ticket resolved with all required information collected
+  failure:
+    description: Max steps reached without resolution
+
+# =========================
+# METADATA
+# =========================
+tags:
+  - customer-support
+  - goal-oriented
+  - information-gathering
+
+difficulty: medium

server7860.py

@@ -0,0 +1,62 @@
+
+# server.py
+
+from fastapi import FastAPI
+from pydantic import BaseModel
+from app.env import CustomerSupportEnv
+
+import json
+
+app = FastAPI()
+
+env = CustomerSupportEnv()
+
+
+class StepRequest(BaseModel):
+    action: str  # OpenEnv sends STRING
+
+
+def parse_action(action_str: str):
+    """
+    Convert string action → dict
+    Supports both:
+    - JSON string
+    - simple commands
+    """
+
+    try:
+        return json.loads(action_str)
+    except:
+        # fallback parsing
+        if action_str == "classify":
+            return {"type": "classify"}
+        elif action_str.startswith("ask_"):
+            return {
+                "type": "ask_info",
+                "field": action_str.replace("ask_", "")
+            }
+        elif action_str == "resolve":
+            return {"type": "resolve"}
+        else:
+            return {"type": "invalid"}
+
+
+@app.post("/reset")
+def reset():
+    obs = env.reset()
+    return obs   # ✅ MUST return raw observation
+
+
+@app.post("/step")
+def step(req: StepRequest):
+
+    action_dict = parse_action(req.action)
+
+    obs, reward, done, info = env.step(action_dict)
+
+    return {
+        "observation": obs,
+        "reward": float(reward),
+        "done": bool(done),
+        "info": info
+    }