Stabilize reset API startup path

app.py

@@ -3,12 +3,12 @@
 from __future__ import annotations
 
 import json
+import os
 from collections import Counter
 from pathlib import Path
 from threading import Lock
 from typing import Any, Dict
 
-import gradio as gr
 from fastapi import FastAPI
 from fastapi.responses import RedirectResponse
 
@@ -21,6 +21,16 @@ if str(PROJECT_ROOT) not in sys.path:
 from environment.env import CodeReviewEnv
 from environment.tasks import TaskDefinitions
 
+ENABLE_GRADIO_UI = os.getenv("ENABLE_GRADIO_UI", "").strip().lower() in {"1", "true", "yes"}
+
+if ENABLE_GRADIO_UI:
+    try:
+        import gradio as gr
+    except Exception:
+        gr = None
+        ENABLE_GRADIO_UI = False
+else:
+    gr = None
 
 app = FastAPI(title="code-review-agent-env")
 _env = CodeReviewEnv()
@@ -502,99 +512,98 @@ body, .gradio-container {
 """
 
 
-task_choices = [t["task_id"] for t in TaskDefinitions.get_all_tasks()]
-
-with gr.Blocks(title="Code Review Agent Environment") as demo:
-    gr.HTML(f"<style>{CUSTOM_CSS}</style>")
-    with gr.Column(elem_classes=["app-shell"]):
-        gr.HTML(
-            """
-            <section class=\"hero\">
-              <h1>Code Review Mission Control</h1>
-              <p>High-clarity operator UI for environment resets, action stepping, and live scoring telemetry.</p>
-              <span class=\"chip mono\">UI: /ui</span>
-              <span class=\"chip mono\">API: /reset /step /state /score /tasks</span>
-              <span class=\"chip mono\">Validation: 3+ graded tasks</span>
-            </section>
-            """
-        )
-
-        with gr.Tabs():
-            with gr.Tab("README"):
-                with gr.Column(elem_id="telemetry-panel"):
-                    gr.Markdown(_readme_markdown())
-                    gr.Markdown(_validation_markdown())
-
-            with gr.Tab("Playground"):
-                with gr.Column(elem_id="control-panel"):
-                    with gr.Row():
-                        task_id_input = gr.Dropdown(choices=task_choices, value=task_choices[0], label="Task ID")
-                        reset_btn = gr.Button("Reset Task", variant="primary")
-                        score_btn = gr.Button("Get Score")
-                        state_btn = gr.Button("Get State")
-
-                    with gr.Row():
-                        score_card = gr.HTML("<div class='metric'><div class='metric-label'>Current Score</div><div class='metric-value'>0.00</div></div>")
-                        step_card = gr.HTML("<div class='metric'><div class='metric-label'>Step</div><div class='metric-value'>0</div></div>")
-                        status_card = gr.HTML("<div class='metric'><div class='metric-label'>Status</div><div class='metric-value'>idle</div></div>")
-
-                    action_input = gr.Textbox(
-                        label="Action JSON",
-                        lines=10,
-                        value=_starter_action_json(task_choices[0]),
-                        elem_classes=["mono"],
-                    )
-                    with gr.Row():
-                        step_btn = gr.Button("Execute Step", variant="primary")
-                        starter_btn = gr.Button("Run Starter Step")
-                        report_btn = gr.Button("Export Episode Report")
-                    gr.Markdown("If you are new, click **Run Starter Step**. It resets the selected task and submits a safe example action.")
-                    output = gr.Code(label="API Response", language="json")
-                    report_out = gr.Code(label="Episode Report", language="json")
-
-            with gr.Tab("Traces"):
-                with gr.Column(elem_id="atlas-panel"):
-                    models, trace_tasks = _trace_choices()
-                    gr.Markdown("### Recorded Traces")
-                    with gr.Row():
-                        trace_model = gr.Dropdown(choices=models, value=models[0], label="Model")
-                        trace_task = gr.Dropdown(choices=trace_tasks, value=trace_tasks[0], label="Task")
-                        trace_refresh = gr.Button("Load Trace")
-                    trace_out = gr.Code(label="Trace Payload", language="json")
-
-            with gr.Tab("Leaderboard"):
-                with gr.Column(elem_id="atlas-panel"):
-                    summary = _benchmark_summary()
-                    gr.Markdown("### Benchmark Leaderboard")
-                    leaderboard_summary = gr.Markdown(f"**Average Task Score:** {summary.get('average_task_score', 0):.3f}  |  **Average Reward:** {summary.get('average_total_reward', 0):.3f}")
-                    leaderboard = gr.Dataframe(
-                        headers=["Rank", "Task", "Task Score", "Total Reward", "Steps", "Model"],
-                        value=_leaderboard_rows(),
-                        interactive=False,
-                        wrap=True,
-                    )
-                    leaderboard_refresh = gr.Button("Refresh Leaderboard")
-
-            with gr.Tab("Tasks"):
-                with gr.Column(elem_id="atlas-panel"):
-                    gr.Markdown("### Task Catalogue")
-                    diff_summary = gr.Textbox(
-                        label="Difficulty Split",
-                        value=_difficulty_summary(),
-                        interactive=False,
-                        elem_classes=["mono"],
-                    )
-                    task_grid = gr.Dataframe(
-                        headers=["Task ID", "Difficulty", "Language", "Name"],
-                        value=_task_table(),
-                        interactive=False,
-                        wrap=True,
-                    )
-                    refresh_tasks_btn = gr.Button("Refresh Task Atlas")
-
-                    task_cards = []
-                    for task in TaskDefinitions.get_all_tasks():
-                        task_cards.append(
+def _build_demo():
+    task_choices = [t["task_id"] for t in TaskDefinitions.get_all_tasks()]
+
+    with gr.Blocks(title="Code Review Agent Environment") as demo:
+        gr.HTML(f"<style>{CUSTOM_CSS}</style>")
+        with gr.Column(elem_classes=["app-shell"]):
+            gr.HTML(
+                """
+                <section class=\"hero\">
+                  <h1>Code Review Mission Control</h1>
+                  <p>High-clarity operator UI for environment resets, action stepping, and live scoring telemetry.</p>
+                  <span class=\"chip mono\">UI: /ui</span>
+                  <span class=\"chip mono\">API: /reset /step /state /score /tasks</span>
+                  <span class=\"chip mono\">Validation: 3+ graded tasks</span>
+                </section>
+                """
+            )
+
+            with gr.Tabs():
+                with gr.Tab("README"):
+                    with gr.Column(elem_id="telemetry-panel"):
+                        gr.Markdown(_readme_markdown())
+                        gr.Markdown(_validation_markdown())
+
+                with gr.Tab("Playground"):
+                    with gr.Column(elem_id="control-panel"):
+                        with gr.Row():
+                            task_id_input = gr.Dropdown(choices=task_choices, value=task_choices[0], label="Task ID")
+                            reset_btn = gr.Button("Reset Task", variant="primary")
+                            score_btn = gr.Button("Get Score")
+                            state_btn = gr.Button("Get State")
+
+                        with gr.Row():
+                            score_card = gr.HTML("<div class='metric'><div class='metric-label'>Current Score</div><div class='metric-value'>0.00</div></div>")
+                            step_card = gr.HTML("<div class='metric'><div class='metric-label'>Step</div><div class='metric-value'>0</div></div>")
+                            status_card = gr.HTML("<div class='metric'><div class='metric-label'>Status</div><div class='metric-value'>idle</div></div>")
+
+                        action_input = gr.Textbox(
+                            label="Action JSON",
+                            lines=10,
+                            value=_starter_action_json(task_choices[0]),
+                            elem_classes=["mono"],
+                        )
+                        with gr.Row():
+                            step_btn = gr.Button("Execute Step", variant="primary")
+                            starter_btn = gr.Button("Run Starter Step")
+                            report_btn = gr.Button("Export Episode Report")
+                        gr.Markdown("If you are new, click **Run Starter Step**. It resets the selected task and submits a safe example action.")
+                        output = gr.Code(label="API Response", language="json")
+                        report_out = gr.Code(label="Episode Report", language="json")
+
+                with gr.Tab("Traces"):
+                    with gr.Column(elem_id="atlas-panel"):
+                        models, trace_tasks = _trace_choices()
+                        gr.Markdown("### Recorded Traces")
+                        with gr.Row():
+                            trace_model = gr.Dropdown(choices=models, value=models[0], label="Model")
+                            trace_task = gr.Dropdown(choices=trace_tasks, value=trace_tasks[0], label="Task")
+                            trace_refresh = gr.Button("Load Trace")
+                        trace_out = gr.Code(label="Trace Payload", language="json")
+
+                with gr.Tab("Leaderboard"):
+                    with gr.Column(elem_id="atlas-panel"):
+                        summary = _benchmark_summary()
+                        gr.Markdown("### Benchmark Leaderboard")
+                        leaderboard_summary = gr.Markdown(f"**Average Task Score:** {summary.get('average_task_score', 0):.3f}  |  **Average Reward:** {summary.get('average_total_reward', 0):.3f}")
+                        leaderboard = gr.Dataframe(
+                            headers=["Rank", "Task", "Task Score", "Total Reward", "Steps", "Model"],
+                            value=_leaderboard_rows(),
+                            interactive=False,
+                            wrap=True,
+                        )
+                        leaderboard_refresh = gr.Button("Refresh Leaderboard")
+
+                with gr.Tab("Tasks"):
+                    with gr.Column(elem_id="atlas-panel"):
+                        gr.Markdown("### Task Catalogue")
+                        diff_summary = gr.Textbox(
+                            label="Difficulty Split",
+                            value=_difficulty_summary(),
+                            interactive=False,
+                            elem_classes=["mono"],
+                        )
+                        task_grid = gr.Dataframe(
+                            headers=["Task ID", "Difficulty", "Language", "Name"],
+                            value=_task_table(),
+                            interactive=False,
+                            wrap=True,
+                        )
+                        refresh_tasks_btn = gr.Button("Refresh Task Atlas")
+
+                        for task in TaskDefinitions.get_all_tasks():
                             gr.Markdown(
                                 f"""
 <div class='task-row'>
@@ -606,45 +615,46 @@ with gr.Blocks(title="Code Review Agent Environment") as demo:
 </div>
                                 """
                             )
-                        )
-
-    def _update_playground_metrics(payload: Dict[str, Any]) -> tuple[str, str, str]:
-        score_value = payload.get("task_score", 0.0)
-        step_value = payload.get("current_step", 0)
-        status_value = "complete" if payload.get("is_complete") else "active"
-        return (
-            f"<div class='metric'><div class='metric-label'>Current Score</div><div class='metric-value'>{float(score_value):.2f}</div></div>",
-            f"<div class='metric'><div class='metric-label'>Step</div><div class='metric-value'>{step_value}</div></div>",
-            f"<div class='metric'><div class='metric-label'>Status</div><div class='metric-value'>{status_value}</div></div>",
-        )
-
-    def _refresh_leaderboard() -> tuple[list[list[str]], str]:
-        summary_data = _benchmark_summary()
-        avg_score = float(summary_data.get("average_task_score", 0.0)) if isinstance(summary_data, dict) else 0.0
-        avg_reward = float(summary_data.get("average_total_reward", 0.0)) if isinstance(summary_data, dict) else 0.0
-        return _leaderboard_rows(), f"### Benchmark Leaderboard\n\n**Average Task Score:** {avg_score:.3f}  |  **Average Reward:** {avg_reward:.3f}"
-
-    def _load_trace(model_name: str, task_id: str) -> str:
-        return _trace_lookup(model_name, task_id)
 
-    reset_btn.click(fn=_ui_reset, inputs=[task_id_input], outputs=[output])
-    step_btn.click(fn=_ui_step, inputs=[action_input], outputs=[output])
-    starter_btn.click(fn=_ui_run_starter_step, inputs=[task_id_input], outputs=[output])
-    state_btn.click(fn=_ui_state, inputs=None, outputs=[output])
-    score_btn.click(fn=_ui_score, inputs=None, outputs=[output])
-    report_btn.click(fn=_episode_report, inputs=None, outputs=[report_out])
-
-    score_btn.click(fn=lambda: _update_playground_metrics(score()), inputs=None, outputs=[score_card, step_card, status_card])
-
-    trace_refresh.click(fn=_load_trace, inputs=[trace_model, trace_task], outputs=[trace_out])
-    leaderboard_refresh.click(fn=_refresh_leaderboard, inputs=None, outputs=[leaderboard, leaderboard_summary])
-    refresh_tasks_btn.click(fn=_difficulty_summary, inputs=None, outputs=[diff_summary])
-    refresh_tasks_btn.click(fn=_task_table, inputs=None, outputs=[task_grid])
+        def _update_playground_metrics(payload: Dict[str, Any]) -> tuple[str, str, str]:
+            score_value = payload.get("task_score", 0.0)
+            step_value = payload.get("current_step", 0)
+            status_value = "complete" if payload.get("is_complete") else "active"
+            return (
+                f"<div class='metric'><div class='metric-label'>Current Score</div><div class='metric-value'>{float(score_value):.2f}</div></div>",
+                f"<div class='metric'><div class='metric-label'>Step</div><div class='metric-value'>{step_value}</div></div>",
+                f"<div class='metric'><div class='metric-label'>Status</div><div class='metric-value'>{status_value}</div></div>",
+            )
+
+        def _refresh_leaderboard() -> tuple[list[list[str]], str]:
+            summary_data = _benchmark_summary()
+            avg_score = float(summary_data.get("average_task_score", 0.0)) if isinstance(summary_data, dict) else 0.0
+            avg_reward = float(summary_data.get("average_total_reward", 0.0)) if isinstance(summary_data, dict) else 0.0
+            return _leaderboard_rows(), f"### Benchmark Leaderboard\n\n**Average Task Score:** {avg_score:.3f}  |  **Average Reward:** {avg_reward:.3f}"
+
+        def _load_trace(model_name: str, task_id: str) -> str:
+            return _trace_lookup(model_name, task_id)
+
+        reset_btn.click(fn=_ui_reset, inputs=[task_id_input], outputs=[output])
+        step_btn.click(fn=_ui_step, inputs=[action_input], outputs=[output])
+        starter_btn.click(fn=_ui_run_starter_step, inputs=[task_id_input], outputs=[output])
+        state_btn.click(fn=_ui_state, inputs=None, outputs=[output])
+        score_btn.click(fn=_ui_score, inputs=None, outputs=[output])
+        report_btn.click(fn=_episode_report, inputs=None, outputs=[report_out])
+        score_btn.click(fn=lambda: _update_playground_metrics(score()), inputs=None, outputs=[score_card, step_card, status_card])
+        trace_refresh.click(fn=_load_trace, inputs=[trace_model, trace_task], outputs=[trace_out])
+        leaderboard_refresh.click(fn=_refresh_leaderboard, inputs=None, outputs=[leaderboard, leaderboard_summary])
+        refresh_tasks_btn.click(fn=_difficulty_summary, inputs=None, outputs=[diff_summary])
+        refresh_tasks_btn.click(fn=_task_table, inputs=None, outputs=[task_grid])
+
+    return demo
 
 
 @app.get("/ui")
 def ui_alias() -> Any:
-    return RedirectResponse(url="/", status_code=307)
-
+    if ENABLE_GRADIO_UI and gr is not None:
+        return RedirectResponse(url="/", status_code=307)
+    return RedirectResponse(url="/docs", status_code=307)
 
-app = gr.mount_gradio_app(app, demo, path="/")
+if ENABLE_GRADIO_UI and gr is not None:
+    app = gr.mount_gradio_app(app, _build_demo(), path="/")

tests/test_server_api.py

@@ -1,5 +1,8 @@
 import unittest
 
+from fastapi.testclient import TestClient
+
+from app import app as fastapi_app
 from server.app import app
 
 
@@ -33,5 +36,23 @@ class TestServerAPI(unittest.TestCase):
         self.assertIn("task_id", payload)
 
 
+class TestFastAPIReset(unittest.TestCase):
+    def setUp(self):
+        self.client = TestClient(fastapi_app)
+
+    def test_post_reset_without_body(self):
+        response = self.client.post("/reset")
+        self.assertEqual(response.status_code, 200)
+        payload = response.json()
+        self.assertIn("observation", payload)
+        self.assertIn("task_description", payload["observation"])
+
+    def test_post_reset_with_task_id_body(self):
+        response = self.client.post("/reset", json={"task_id": "bug_detection_easy_1"})
+        self.assertEqual(response.status_code, 200)
+        payload = response.json()
+        self.assertEqual(payload["observation"]["task_difficulty"], "easy")
+
+
 if __name__ == "__main__":
     unittest.main()