Protokoll #15130

{
    "event": "PreToolUse",
    "tool_name": "Read",
    "tool_input": {
        "file_path": "\/var\/www\/scripts\/pipeline\/generate.py"
    }
}

{
    "tool_response": {
        "type": "text",
        "file": {
            "filePath": "\/var\/www\/scripts\/pipeline\/generate.py",
            "content": "#!\/usr\/bin\/env python3\n\"\"\"\nContent Generation for KI-System\nGenerates content using RAG context, author profiles, and contracts.\n\"\"\"\n\nimport json\nimport sys\n\nsys.path.insert(0, \"\/var\/www\/scripts\/pipeline\")\n\nfrom config import ANTHROPIC_API_KEY, ANTHROPIC_MODEL, OLLAMA_CHAT_MODEL, OLLAMA_HOST  # noqa: I001, E402\nfrom db import db  # noqa: E402\nfrom embed import search_similar  # noqa: E402\n\n\ndef _repair_json(json_str):\n    \"\"\"\n    Attempt to repair common JSON issues from LLM output.\n\n    Fixes:\n    - Unescaped quotes in strings\n    - Missing commas between array elements\n    - Trailing commas\n    - Control characters in strings\n    \"\"\"\n    import re\n\n    # Remove control characters except newlines and tabs\n    json_str = re.sub(r\"[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f]\", \"\", json_str)\n\n    # Fix common issue: missing comma before closing bracket in arrays\n    json_str = re.sub(r'\"\\s*\\n\\s*]', '\"\\n]', json_str)\n\n    # Fix trailing commas before closing brackets\/braces\n    json_str = re.sub(r\",\\s*}\", \"}\", json_str)\n    json_str = re.sub(r\",\\s*]\", \"]\", json_str)\n\n    # Fix missing commas between array elements (string followed by string)\n    json_str = re.sub(r'\"\\s*\\n\\s*\"', '\",\\n\"', json_str)\n\n    # Fix unescaped quotes within strings (heuristic: quotes not at boundaries)\n    # This is tricky, so we do a simple fix for common patterns\n    lines = json_str.split(\"\\n\")\n    fixed_lines = []\n    for line in lines:\n        # Count quotes - if odd number and line has content, try to fix\n        quote_count = line.count('\"') - line.count('\\\\\"')\n        if quote_count % 2 != 0 and \":\" in line:\n            # Try to escape internal quotes (very basic heuristic)\n            parts = line.split(\":\", 1)\n            if len(parts) == 2:\n                key_part = parts[0]\n                value_part = parts[1]\n                # If value has odd quotes, try to balance\n                if value_part.count('\"') % 2 != 0:\n                    # Add escaped quote or remove problematic char\n                    value_part = value_part.rstrip().rstrip(\",\")\n                    if not value_part.endswith('\"'):\n                        value_part += '\"'\n                    line = key_part + \":\" + value_part\n        fixed_lines.append(line)\n\n    return \"\\n\".join(fixed_lines)\n\n\ndef get_prompt(name):\n    \"\"\"Load prompt from database by name.\"\"\"\n    cursor = db.execute(\n        \"\"\"SELECT content FROM prompts WHERE name = %s AND is_active = 1 ORDER BY version DESC LIMIT 1\"\"\",\n        (name,),\n    )\n    result = cursor.fetchone()\n    cursor.close()\n    return result[\"content\"] if result else None\n\n\ndef get_rag_context(briefing, collection=\"documents\", limit=5):\n    \"\"\"\n    Get relevant context from Qdrant based on briefing.\n    Returns list of chunks with content and metadata.\n    \"\"\"\n    results = search_similar(briefing, collection=collection, limit=limit)\n\n    context_items = []\n    for result in results:\n        context_items.append(\n            {\n                \"content\": result[\"payload\"].get(\"content\", \"\"),\n                \"source\": result[\"payload\"].get(\"document_title\", \"Unknown\"),\n                \"score\": round(result[\"score\"], 4),\n                \"chunk_id\": result[\"payload\"].get(\"chunk_id\"),\n                \"document_id\": result[\"payload\"].get(\"document_id\"),\n            }\n        )\n\n    return context_items\n\n\ndef get_config_item(item_id, item_type):\n    \"\"\"Load configuration item from content_config table.\"\"\"\n    if not item_id:\n        return None\n\n    cursor = db.execute(\n        \"SELECT name, content FROM content_config WHERE id = %s AND type = %s AND status = 'active'\",\n        (item_id, item_type),\n    )\n    result = cursor.fetchone()\n    cursor.close()\n\n    if result:\n        config = json.loads(result[\"content\"]) if isinstance(result[\"content\"], str) else result[\"content\"]\n        return {\"name\": result[\"name\"], \"config\": config}\n    return None\n\n\ndef get_semantic_context(chunk_ids):\n    \"\"\"\n    Load entities and relations based on chunk_ids.\n\n    Uses the chunk_entities junction table to find relevant entities,\n    then loads relations between those entities.\n\n    Args:\n        chunk_ids: List of chunk IDs from RAG context\n\n    Returns:\n        dict with 'entities' and 'relations' lists\n    \"\"\"\n    if not chunk_ids:\n        return {\"entities\": [], \"relations\": []}\n\n    # Filter out None values\n    chunk_ids = [cid for cid in chunk_ids if cid is not None]\n    if not chunk_ids:\n        return {\"entities\": [], \"relations\": []}\n\n    placeholders = \", \".join([\"%s\"] * len(chunk_ids))\n\n    # Load entities via chunk_entities\n    cursor = db.execute(\n        f\"\"\"SELECT DISTINCT e.id, e.name, e.type, e.description,\n                  AVG(ce.relevance_score) as relevance\n           FROM chunk_entities ce\n           JOIN entities e ON ce.entity_id = e.id\n           WHERE ce.chunk_id IN ({placeholders})\n           GROUP BY e.id, e.name, e.type, e.description\n           ORDER BY relevance DESC\n           LIMIT 10\"\"\",\n        tuple(chunk_ids),\n    )\n    entities = cursor.fetchall()\n    cursor.close()\n\n    if not entities:\n        return {\"entities\": [], \"relations\": []}\n\n    # Get entity IDs for relation lookup\n    entity_ids = [e[\"id\"] for e in entities]\n    entity_placeholders = \", \".join([\"%s\"] * len(entity_ids))\n\n    # Load relations between found entities\n    cursor = db.execute(\n        f\"\"\"SELECT e1.name as source, er.relation_type, e2.name as target\n           FROM entity_relations er\n           JOIN entities e1 ON er.source_entity_id = e1.id\n           JOIN entities e2 ON er.target_entity_id = e2.id\n           WHERE e1.id IN ({entity_placeholders}) AND e2.id IN ({entity_placeholders})\n           LIMIT 15\"\"\",\n        tuple(entity_ids) + tuple(entity_ids),\n    )\n    relations = cursor.fetchall()\n    cursor.close()\n\n    return {\"entities\": entities, \"relations\": relations}\n\n\ndef get_taxonomy_context(document_ids):\n    \"\"\"\n    Load taxonomy terms for documents.\n\n    Args:\n        document_ids: List of document IDs from RAG context\n\n    Returns:\n        List of taxonomy term dicts with name, slug, confidence\n    \"\"\"\n    if not document_ids:\n        return []\n\n    # Filter out None values\n    document_ids = [did for did in document_ids if did is not None]\n    if not document_ids:\n        return []\n\n    placeholders = \", \".join([\"%s\"] * len(document_ids))\n\n    cursor = db.execute(\n        f\"\"\"SELECT DISTINCT tt.name, tt.slug, MAX(dt.confidence) as confidence\n           FROM document_taxonomy dt\n           JOIN taxonomy_terms tt ON dt.taxonomy_term_id = tt.id\n           WHERE dt.document_id IN ({placeholders})\n           GROUP BY tt.id, tt.name, tt.slug\n           ORDER BY confidence DESC\"\"\",\n        tuple(document_ids),\n    )\n    taxonomy = cursor.fetchall()\n    cursor.close()\n\n    return taxonomy\n\n\ndef get_author_profile(profile_id):\n    \"\"\"Load author profile from database.\"\"\"\n    return get_config_item(profile_id, \"author_profile\")\n\n\ndef get_contract(contract_id):\n    \"\"\"Load content contract from database.\"\"\"\n    return get_config_item(contract_id, \"contract\")\n\n\ndef get_structure(structure_id):\n    \"\"\"Load content structure from database.\"\"\"\n    result = get_config_item(structure_id, \"structure\")\n    if result:\n        # Structure has additional 'type' field in config\n        result[\"type\"] = result[\"config\"].get(\"type\", \"article\")\n    return result\n\n\ndef get_order(order_id):\n    \"\"\"Load content order with all related data.\"\"\"\n    cursor = db.execute(\n        \"\"\"SELECT co.*,\n                  ap.name as profile_name, ap.content as profile_config,\n                  cc.name as contract_name, cc.content as contract_config,\n                  cs.name as structure_name, cs.content as structure_config\n           FROM content_orders co\n           LEFT JOIN content_config ap ON co.author_profile_id = ap.id AND ap.type = 'author_profile'\n           LEFT JOIN content_config cc ON co.contract_id = cc.id AND cc.type = 'contract'\n           LEFT JOIN content_config cs ON co.structure_id = cs.id AND cs.type = 'structure'\n           WHERE co.id = %s\"\"\",\n        (order_id,),\n    )\n    result = cursor.fetchone()\n    cursor.close()\n    return result\n\n\ndef _parse_new_author_profile(config):\n    \"\"\"Parse new-style author profile (Cary format) into prompt text.\"\"\"\n    sections = []\n\n    # Haltung\n    haltung = config.get(\"haltung\", {})\n    if haltung:\n        sections.append(f\"\"\"### Haltung:\n- Grundhaltung: {haltung.get(\"grundhaltung\", \"\")}\n- Ausrichtung: {haltung.get(\"ausrichtung\", \"\")}\n- Spannungstoleranz: {haltung.get(\"spannungstoleranz\", \"\")}\n- Vereinfachung: {haltung.get(\"vereinfachung\", \"\")}\"\"\")\n\n    # Tonalität\n    tonalitaet = config.get(\"tonalitaet\", {})\n    if tonalitaet:\n        sections.append(f\"\"\"### Tonalität:\n- Charakter: {tonalitaet.get(\"charakter\", \"\")}\n- Stil: {tonalitaet.get(\"stil\", \"\")}\n- Wirkung: {tonalitaet.get(\"wirkung\", \"\")}\n- Abgrenzung: {tonalitaet.get(\"abgrenzung\", \"\")}\"\"\")\n\n    # Sprachmodus\n    sprachmodus = config.get(\"sprachmodus\", {})\n    if sprachmodus:\n        sections.append(f\"\"\"### Sprachmodus:\n- Denkstil: {sprachmodus.get(\"denkstil\", \"\")}\n- Aussagenform: {sprachmodus.get(\"aussagenform\", \"\")}\n- Fragenfunktion: {sprachmodus.get(\"fragenfunktion\", \"\")}\"\"\")\n\n    # Grammatik und Satzbau - WICHTIG für Verbote\n    grammatik = config.get(\"grammatik_und_satzbau\", {})\n    if grammatik:\n        verbote = []\n        if grammatik.get(\"stakkato\") == \"ausgeschlossen\":\n            verbote.append(\"Stakkato-Sätze\")\n        if grammatik.get(\"einschuebe\") == \"keine\":\n            verbote.append(\"Einschübe\")\n        if grammatik.get(\"gedankenstriche\") == \"verboten\":\n            verbote.append(\"Gedankenstriche (–)\")\n\n        sections.append(f\"\"\"### Grammatik und Satzbau:\n- Sätze: {grammatik.get(\"saetze\", \"\")}\n- Rhythmus: {grammatik.get(\"rhythmus\", \"\")}\n- **VERBOTEN:** {\", \".join(verbote) if verbote else \"keine\"}\"\"\")\n\n    # Wortwahl\n    wortwahl = config.get(\"wortwahl\", {})\n    if wortwahl:\n        verboten = []\n        if wortwahl.get(\"buzzwords\") == \"ausge... [TRUNCATED-2e4bdc5f812ef488]",
            "numLines": 1133,
            "startLine": 1,
            "totalLines": 1133
        }
    }
}