lesson-model-output-fix.json

{
  "task_id": "lesson-model-output-fix",
  "title": "Model Output Fix",
  "domain": "general",
  "tags": [
    "contrib"
  ],
  "problem": "模型返回的内容不完整（truncated），或者 JSON 解析失败（`json.decoder.JSONDecodeError`），导致后续处理流程中断。",
  "solution": "**JSON 截断修复：**\n```python\nimport json, re\n\ndef safe_parse_json(raw: str) -> dict | None:\n    \"\"\"尝试修复截断的 JSON\"\"\"\n    # 1. 去除控制字符\n    raw = re.sub(r'[\\x00-\\x1f]', '', raw)\n    # 2. 尝试直接解析\n    try:\n        return json.loads(raw)\n    except json.JSONDecodeError:\n        pass\n    # 3. 尝试补全缺失的闭合括号\n    for bracket in [('\"', '\"'), (\"'\", \"'\"), (\"[\", \"]\"), (\"{\", \"}\")]:\n        open_count = raw.count(bracket[0]) - raw.count(bracket[1])\n        if open_count > 0:\n            raw += bracket[1] * open_count\n    try:\n        return json.loads(raw)\n    except json.JSONDecodeError:\n        return None\n```\n\n**模型侧预防：**\n```python\n# 增加 max_tokens 或改用更强的模型\nresponse = client.messages.create(\n    model=\"claude-sonnet-4-6\",\n    max_tokens=4096,  # 加大\n    # 或者在 prompt 里要求：\"输出完整的 JSON，不要省略任何字段\"\n)\n```\n\n**检测截断：**\n```python\ndef is_truncated(response_text: str) -> bool:\n    \"\"\"检查是否被截断\"\"\"\n    # 常见截断标记\n    truncated_markers = [\n        '...',           # 省略号结尾\n        '...\"',          # 字符串未闭合\n        '{...}',         # 对象被省略\n        '\"',             # 字符串未闭合\n    ]\n    for marker in truncated_markers:\n        if response_text.rstrip().endswith(marker):\n            return True\n    return False\n```",
  "source": "lessons/contrib/model-output-fix.md",
  "test_cmd": ""
}