Protokoll #24738

{
    "event": "PostToolUse",
    "tool_name": "Read",
    "tool_response": {
        "type": "text",
        "file": {
            "filePath": "\/var\/www\/scripts\/pipeline\/enrich.py",
            "content": "#!\/usr\/bin\/env python3\n\"\"\"\nVision-Enrichment module for KI-System Pipeline.\nEnriches chunks with visual context from page-level vision analysis.\n\nUsage:\n    python enrich.py <document_id>     # Enrich all chunks for a document\n    python enrich.py --all             # Enrich all documents with vision data\n\"\"\"\n\nimport json\nimport re\nimport sys\n\nfrom db import db\n\n\ndef get_vision_context(document_id, page_number):\n    \"\"\"\n    Get vision analysis for a specific page.\n\n    Returns dict with structured vision info or None.\n    \"\"\"\n    cursor = db.execute(\n        \"\"\"SELECT vision_analysis\n           FROM document_pages\n           WHERE document_id = %s AND page_number = %s\"\"\",\n        (document_id, page_number),\n    )\n    result = cursor.fetchone()\n    cursor.close()\n\n    if not result or not result.get(\"vision_analysis\"):\n        return None\n\n    try:\n        vision_data = json.loads(result[\"vision_analysis\"])\n        return vision_data\n    except (json.JSONDecodeError, TypeError):\n        return None\n\n\ndef extract_vision_summary(vision_data):\n    \"\"\"\n    Extract key information from vision analysis for chunk enrichment.\n\n    Returns compact dict with:\n    - detected_elements: list of visual elements found\n    - page_title: extracted title if any\n    - has_images: bool\n    - has_charts: bool\n    - has_tables: bool\n    - layout_type: detected layout style\n    - key_topics: extracted key topics\/concepts\n    \"\"\"\n    if not vision_data:\n        return None\n\n    analysis_text = vision_data.get(\"analysis\", \"\")\n    if not analysis_text:\n        return None\n\n    summary = {\n        \"detected_elements\": [],\n        \"page_title\": None,\n        \"has_images\": False,\n        \"has_charts\": False,\n        \"has_tables\": False,\n        \"layout_type\": \"standard\",\n        \"key_topics\": [],\n        \"vision_tokens\": vision_data.get(\"tokens\", 0),\n    }\n\n    # Detect visual elements\n    analysis_lower = analysis_text.lower()\n\n    # Check for images\n    if any(word in analysis_lower for word in [\"bild\", \"foto\", \"image\", \"abbildung\", \"grafik\"]):  # noqa: SIM102\n        if \"keine\" not in analysis_lower.split(\"bild\")[0][-20:] if \"bild\" in analysis_lower else True:\n            summary[\"has_images\"] = True\n            summary[\"detected_elements\"].append(\"images\")\n\n    # Check for charts\/diagrams\n    if any(word in analysis_lower for word in [\"diagramm\", \"chart\", \"graph\", \"schaubild\"]):  # noqa: SIM102\n        if \"keine\" not in analysis_lower.split(\"diagramm\")[0][-20:] if \"diagramm\" in analysis_lower else True:\n            summary[\"has_charts\"] = True\n            summary[\"detected_elements\"].append(\"charts\")\n\n    # Check for tables\n    if any(word in analysis_lower for word in [\"tabelle\", \"table\", \"übersicht\"]):  # noqa: SIM102\n        if \"keine\" not in analysis_lower.split(\"tabelle\")[0][-20:] if \"tabelle\" in analysis_lower else True:\n            summary[\"has_tables\"] = True\n            summary[\"detected_elements\"].append(\"tables\")\n\n    # Check for callouts\/highlights\n    if any(word in analysis_lower for word in [\"callout\", \"hervorhebung\", \"box\", \"kasten\", \"zitat\"]):\n        summary[\"detected_elements\"].append(\"callouts\")\n\n    # Extract title (look for patterns like \"Titel: X\" or \"Überschrift: X\")\n    title_patterns = [\n        r'[\"\\']([^\"\\']{5,60})[\"\\']',  # Quoted strings\n        r'Titel[:\\s]+[\"\\']?([^\"\\'\\n]{5,60})',\n        r'Überschrift[:\\s]+[\"\\']?([^\"\\'\\n]{5,60})',\n    ]\n\n    for pattern in title_patterns:\n        match = re.search(pattern, analysis_text)\n        if match:\n            potential_title = match.group(1).strip()\n            # Filter out common non-titles\n            if not any(\n                skip in potential_title.lower() for skip in [\"keine\", \"nicht\", \"gibt es\", \"vorhanden\", \"enthält\"]\n            ):\n                summary[\"page_title\"] = potential_title[:100]\n                break\n\n    # Detect layout type\n    if any(word in analysis_lower for word in [\"zwei spalten\", \"zweispaltig\", \"columns\"]):\n        summary[\"layout_type\"] = \"two-column\"\n    elif any(word in analysis_lower for word in [\"liste\", \"aufzählung\", \"bullet\"]):\n        summary[\"layout_type\"] = \"list\"\n    elif any(word in analysis_lower for word in [\"vollbild\", \"full page\", \"ganzseitig\"]):\n        summary[\"layout_type\"] = \"full-page\"\n\n    # Extract key topics (look for bold\/emphasized terms)\n    bold_pattern = r\"\\*\\*([^*]+)\\*\\*\"\n    bold_matches = re.findall(bold_pattern, analysis_text)\n    if bold_matches:\n        # Filter and dedupe\n        topics = []",
            "numLines": 130,
            "startLine": 1,
            "totalLines": 298
        }
    }
}

-