Upload 5 files

README.md

@@ -1,3 +1,43 @@
----
-license: mit
----
+# QuestBench
+
+**QuestBench** is a course-curated benchmark of **256 expert-level questions**
+for evaluating the deep-search capabilities of language models. Questions are
+long-tail queries spanning **14 normalized domains** in the humanities and
+social sciences (Literature, Law, History, Foreign Languages, Social
+Sciences, Arts, Archaeology, Journalism, …). Each question, reference answer,
+and grading rubric is in Chinese; the model under evaluation may search and
+read in any language.
+
+## Files
+
+- `questbench.jsonl` — 256 questions, one JSON record per line.
+- `.claude.json` — [Claude Code](https://www.anthropic.com/claude-code) MCP server config.
+- `vision_server.py` — local MCP server proxying image analysis to OpenRouter.
+- `package.json`, `package-lock.json` — npm dependencies.
+
+Each JSONL record has four fields: `domain`, `question`, `answer`,
+`grading_criteria` (the rubric an LLM judge uses to assign a 0–100 score).
+
+## Run
+
+Setup:
+
+```bash
+brew install --cask claude        # Claude Code CLI
+brew install jq node uv
+npm install
+export ANTHROPIC_API_KEY=...      # Claude
+export SERPER_API_KEY=...         # https://serper.dev
+export OPENROUTER_API_KEY=...     # https://openrouter.ai (vision)
+```
+
+Run a single question:
+
+```bash
+QUESTION=$(jq -r '.question' questbench.jsonl | head -n 1) && \
+  claude -p "$QUESTION" \
+    --mcp-config ./.claude.json --strict-mcp-config \
+    --permission-mode bypassPermissions \
+    --disallowedTools WebSearch WebFetch \
+    --output-format stream-json --verbose
+```

package-lock.json

@@ -0,0 +1,63 @@
+{
+  "name": "dr-bench",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {
+    "": {
+      "dependencies": {
+        "claude": "^0.1.1",
+        "playwright": "^1.59.1"
+      }
+    },
+    "node_modules/claude": {
+      "version": "0.1.1",
+      "resolved": "https://registry.npmjs.org/claude/-/claude-0.1.1.tgz",
+      "integrity": "sha512-j7oSibqQdIODNhkI1sEJzHMiPsF43L/GqNbcA+eDDyGM10+x2sH9NW/PK6vM3z0J2tLDKMBcc5ZjVaoRinhuCA==",
+      "license": "ISC"
+    },
+    "node_modules/fsevents": {
+      "version": "2.3.2",
+      "resolved": "https://registry.npmjs.org/fsevents/-/fsevents-2.3.2.tgz",
+      "integrity": "sha512-xiqMQR4xAeHTuB9uWm+fFRcIOgKBMiOBP+eXiyT7jsgVCq1bkVygt00oASowB7EdtpOHaaPgKt812P9ab+DDKA==",
+      "hasInstallScript": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": "^8.16.0 || ^10.6.0 || >=11.0.0"
+      }
+    },
+    "node_modules/playwright": {
+      "version": "1.59.1",
+      "resolved": "https://registry.npmjs.org/playwright/-/playwright-1.59.1.tgz",
+      "integrity": "sha512-C8oWjPR3F81yljW9o5OxcWzfh6avkVwDD2VYdwIGqTkl+OGFISgypqzfu7dOe4QNLL2aqcWBmI3PMtLIK233lw==",
+      "license": "Apache-2.0",
+      "dependencies": {
+        "playwright-core": "1.59.1"
+      },
+      "bin": {
+        "playwright": "cli.js"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "optionalDependencies": {
+        "fsevents": "2.3.2"
+      }
+    },
+    "node_modules/playwright-core": {
+      "version": "1.59.1",
+      "resolved": "https://registry.npmjs.org/playwright-core/-/playwright-core-1.59.1.tgz",
+      "integrity": "sha512-HBV/RJg81z5BiiZ9yPzIiClYV/QMsDCKUyogwH9p3MCP6IYjUFu/MActgYAvK0oWyV9NlwM3GLBjADyWgydVyg==",
+      "license": "Apache-2.0",
+      "bin": {
+        "playwright-core": "cli.js"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    }
+  }
+}

package.json

@@ -0,0 +1,6 @@
+{
+  "dependencies": {
+    "claude": "^0.1.1",
+    "playwright": "^1.59.1"
+  }
+}

vision_server.py

@@ -0,0 +1,122 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["mcp>=1.0", "openai>=1.0", "requests>=2.31"]
+# ///
+import os
+import base64
+import tempfile
+import requests
+from urllib.parse import urlparse
+from mcp.server.fastmcp import FastMCP
+from openai import OpenAI
+
+mcp = FastMCP("Robust-Vision-Server")
+
+
+def is_valid_url(url: str) -> bool:
+    """Return True iff the input is a syntactically valid http/https URL."""
+    try:
+        result = urlparse(url)
+        return all([result.scheme in ['http', 'https'], result.netloc])
+    except ValueError:
+        return False
+
+
+def get_base64_from_local_file(file_path: str) -> str:
+    """Read a local image file and encode it as a base64 data URI."""
+    if not os.path.exists(file_path):
+        raise FileNotFoundError(f"local file not found: {file_path}")
+
+    with open(file_path, "rb") as f:
+        encoded = base64.b64encode(f.read()).decode('utf-8')
+
+    # OpenRouter accepts a generic image/jpeg URI; the API infers the real format.
+    return f"data:image/jpeg;base64,{encoded}"
+
+
+@mcp.tool()
+def analyze_image_with_openrouter(
+    image_source: str,
+    prompt: str = "Describe the image in detail and extract any salient text or information.",
+    model_name: str = "bytedance-seed/seed-2.0-lite",
+) -> str:
+    """
+    Analyze an image with an OpenRouter vision model. Accepts both local file
+    paths and remote http/https URLs. Remote images are downloaded to a
+    temporary directory before being sent to the API.
+
+    Args:
+        image_source: absolute local path or http/https URL.
+        prompt: instruction passed to the vision model.
+        model_name: OpenRouter model id.
+    """
+    api_key = os.getenv("OPENROUTER_API_KEY")
+    if not api_key:
+        return "Error: OPENROUTER_API_KEY is not set."
+
+    client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=api_key)
+
+    # Branch A: remote URL — download to a temp dir, then call the API.
+    if is_valid_url(image_source):
+        with tempfile.TemporaryDirectory() as temp_dir:
+            temp_file_path = os.path.join(temp_dir, "downloaded_image.tmp")
+
+            try:
+                headers = {
+                    "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36",
+                    "Accept": "image/webp,image/apng,image/*,*/*;q=0.8",
+                }
+                response = requests.get(image_source, headers=headers, stream=True, timeout=15)
+                response.raise_for_status()
+
+                content_type = response.headers.get('Content-Type', '')
+                if not content_type.startswith('image/'):
+                    return f"Download failed: target URL did not return an image (Content-Type: {content_type})."
+
+                with open(temp_file_path, 'wb') as f:
+                    for chunk in response.iter_content(chunk_size=8192):
+                        f.write(chunk)
+
+                image_data_uri = get_base64_from_local_file(temp_file_path)
+
+            except requests.exceptions.Timeout:
+                return "Download failed: request timed out (>15s)."
+            except requests.exceptions.HTTPError as err:
+                return f"Download failed: HTTP error {err.response.status_code}."
+            except requests.exceptions.RequestException as err:
+                return f"Download failed: network error: {err}"
+            except Exception as err:
+                return f"Temp-file processing failed: {err}"
+
+            # Call inside the `with` block so the temp file outlives the request.
+            return _call_openrouter_api(client, model_name, prompt, image_data_uri)
+
+    # Branch B: local file path.
+    try:
+        image_data_uri = get_base64_from_local_file(image_source)
+        return _call_openrouter_api(client, model_name, prompt, image_data_uri)
+    except Exception as err:
+        return f"Local file processing failed: {err}"
+
+
+def _call_openrouter_api(client: OpenAI, model: str, prompt: str, image_data_uri: str) -> str:
+    try:
+        response = client.chat.completions.create(
+            model=model,
+            messages=[
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": prompt},
+                        {"type": "image_url", "image_url": {"url": image_data_uri}},
+                    ],
+                }
+            ],
+        )
+        return response.choices[0].message.content
+    except Exception as err:
+        return f"OpenRouter API call failed: {err}"
+
+
+if __name__ == "__main__":
+    mcp.run()