Update openenv.yaml

openenv.yaml

@@ -7,27 +7,61 @@ port: 7860
 description: "Carbon-aware RL environment for multi-objective LLM query routing"
 version: "1.0.0"
 
+api:
+  base_url: /
+  endpoints:
+    reset:
+      method: POST
+      path: /reset
+    step:
+      method: POST
+      path: /step
+    state:
+      method: GET
+      path: /state
+    tasks:
+      method: GET
+      path: /tasks
+    grader:
+      method: POST
+      path: /grader
+
 tasks:
-  - id: "easy"
-    grader: "graders:EasyGrader"
-    rubric: "graders:EasyGrader"
-    name: "Single Query Routing"
-    title: "Single Query Routing"
-    description: "Route a single LLM query to the optimal model tier while minimising carbon footprint and latency. The agent must choose a model that meets the query minimum capability requirement."
-    difficulty: "easy"
+  - id: task_1
+    task_id: task_1
+    name: Single Query Routing
+    difficulty: easy
+    description: "Route a single LLM query to the optimal model tier while minimising carbon footprint and latency."
+    max_steps: 10
+    grader: graders:grade_task_1
+    graders:
+      - graders:grade_task_1
+    reward_range:
+      - 0.0
+      - 1.0
 
-  - id: "medium"
-    grader: "graders:MediumGrader"
-    rubric: "graders:MediumGrader"
-    name: "Multi-Query Episode"
-    title: "Multi-Query Episode"
-    description: "Route three queries of varying complexity. An extra penalty applies when the LARGE model is used unnecessarily. The agent must balance accuracy against energy cost over the 3-query episode."
-    difficulty: "medium"
+  - id: task_2
+    task_id: task_2
+    name: Multi-Query Episode
+    difficulty: medium
+    description: "Route 3 queries of varying complexity. LARGE model incurs -0.2 penalty per use."
+    max_steps: 20
+    grader: graders:grade_task_2
+    graders:
+      - graders:grade_task_2
+    reward_range:
+      - 0.0
+      - 1.0
 
-  - id: "hard"
-    grader: "graders:HardGrader"
-    rubric: "graders:HardGrader"
-    name: "Stateful Carbon-Aware Routing"
-    title: "Stateful Carbon-Aware Routing"
-    description: "Route five queries including repeated queries, a KB-answerable question, and a compliance query requiring LARGE. The agent must exploit caching, cascade strategies, KB lookups, and WAIT actions to maximise cumulative reward."
-    difficulty: "hard"
+  - id: task_3
+    task_id: task_3
+    name: Stateful Carbon-Aware Routing
+    difficulty: hard
+    description: "5-query stateful episode with caching, KB lookups, cascade strategies, and carbon-aware waiting."
+    max_steps: 50
+    grader: graders:grade_task_3
+    graders:
+      - graders:grade_task_3
+    reward_range:
+      - 0.0
+      - 1.0