Upload folder using huggingface_hub

server/app.py

@@ -1,6 +1,6 @@
 """
-FastAPI server wrapping CodeReviewEnv-RX.
-Exposes OpenEnv-compliant endpoints: /reset, /step, /state, /grader, /tasks, /baseline, /health
+FastAPI server wrapping DebugOps-RX.
+Exposes OpenEnv-compliant endpoints: /reset, /step, /state, /grader, /health
 """
 
 from fastapi import FastAPI, HTTPException, Query
@@ -9,27 +9,23 @@ from contextlib import asynccontextmanager
 import sys
 import os
 
-# Add parent directory to path so we can import models/tasks
+# Add parent directory to path
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-from models import Action, TaskInfo  # pyre-ignore
-from server.env import CodeReviewEnv  # pyre-ignore
-from tasks import get_tasks  # pyre-ignore
+from models import Action  # pyre-ignore
+from server.env import DebugOpsEnv  # pyre-ignore
 
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
-    """Initialize default environment on startup."""
     app.state.envs = {}
     yield
 
 
 app = FastAPI(
-    title="CodeReviewEnv-RX",
-    description="A research-grade, interactive, step-based environment where AI agents "
-                "navigate multi-file codebases, inspect lines, trace function calls, "
-                "and flag issues under time and step constraints.",
-    version="2.0.0",
+    title="DebugOps-RX API",
+    description="Realistic eXecution benchmark for debugging agents.",
+    version="1.0.0",
     lifespan=lifespan,
 )
 
@@ -41,95 +37,56 @@ app.add_middleware(
 )
 
 
-def _get_env() -> CodeReviewEnv:
-    """Get the current environment instance or raise 400."""
+def _get_env() -> DebugOpsEnv:
     env = app.state.envs.get("current")
     if env is None:
         raise HTTPException(status_code=400, detail="No active environment. Call /reset first.")
     return env
 
 
-# ── Endpoints ──────────────────────────────────────────
-
-
 @app.post("/reset")
 def reset(
-    task_level: str = Query("easy", description="Task level: easy, medium, hard, security, concurrency, hard_rx"),
-    seed: int = Query(42, description="Random seed for deterministic episodes"),
+    difficulty: str = Query("easy", description="easy, medium, hard, extreme"),
+    split: str = Query("test", description="train, test, ood"),
+    seed: int = Query(42),
 ):
-    """Reset the environment with a specific task and seed. Returns the first observation."""
-    valid_levels = list(get_tasks().keys())
-    if task_level not in valid_levels:
-        raise HTTPException(
-            status_code=400,
-            detail=f"Invalid task_level: {task_level}. Must be one of {valid_levels}"
-        )
-    env = CodeReviewEnv(task_level=task_level, seed=seed)
+    env = DebugOpsEnv(seed=seed)
     app.state.envs["current"] = env
-    obs = env.reset()
+    obs = env.reset(difficulty=difficulty, split=split)
     return obs.model_dump()
 
 
 @app.post("/step")
 def step(action: Action):
-    """Submit an action (open_file, inspect, flag, trace). Returns observation + reward."""
     env = _get_env()
     try:
-        obs, reward = env.step(action)
+        obs, reward, done, info = env.step(action)
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
     return {
         "observation": obs.model_dump(),
-        "reward": reward.model_dump(),
+        "reward": reward,
+        "done": done,
+        "info": info
     }
 
 
 @app.get("/state")
 def state():
-    """Return full internal state of the current environment."""
     env = _get_env()
-    return env.state()
+    return env.state.model_dump()
 
 
 @app.get("/grader")
 def grader():
-    """Grade the current episode. Returns deterministic score 0.0–1.0 with breakdown."""
     env = _get_env()
-    result = env.grade_episode()
-    return result.model_dump()
-
-
-@app.get("/tasks")
-def list_tasks():
-    """List all available tasks with metadata."""
-    tasks = get_tasks()
-    task_list = []
-    for level, task in tasks.items():
-        task_list.append(
-            TaskInfo(
-                name=task.name,
-                level=task.level,
-                description=task.description,
-                num_files=len(task.files),
-                num_required_issues=len([i for i in task.issues if i.confidence >= 0.5]),
-                num_optional_issues=len(task.optional_issues),
-            ).model_dump()
-        )
-    return {"tasks": task_list}
-
-
-@app.post("/baseline")
-def run_baseline():
-    """
-    Run a deterministic heuristic baseline across all tasks.
-    Returns scores per task level.
-    """
-    from baseline import run_heuristic_baseline  # pyre-ignore
-    scores = run_heuristic_baseline()
-    return {"baseline_scores": scores}
+    score = env.grade(env.trajectory)
+    return {
+        "score": score.final(),
+        "breakdown": score.model_dump()
+    }
 
 
 @app.get("/health")
 def health():
-    """Health check endpoint."""
-    return {"status": "ok", "environment": "CodeReviewEnv-RX", "version": "2.0.0"}
+    return {"status": "ok", "environment": "DebugOps-RX", "version": "1.0.0"}