DeepRead 税务表格解析

将 W-2、1099-NEC/MISC/INT/DIV、1040 等各类税务文件转化为结构清晰、格式规范的 JSON 数据——涵盖雇主/付款方信息、收款方信息，以及逐项税表栏位（box-by-box）金额，并为每个字段标注 needs_review 标志，确保任何错误均无法流入报税流程或收入计算环节。

本技能指示 agent 向 https://api.deepread.tech 发送 POST 请求上传文档，并轮询获取处理结果。不修改任何系统文件。

返回结果示例（W-2）

{
  "schema_version": "dp02",
  "status": "completed",
  "extraction": {
    "fields": [
      {"key": "form_type", "value": "W-2", "needs_review": false, "location": {"page": 1}},
      {"key": "tax_year", "value": 2025, "needs_review": false, "location": {"page": 1}},
      {"key": "employer_name", "value": "Acme Corp", "needs_review": false, "location": {"page": 1}},
      {"key": "employer_ein", "value": "12-3456789", "needs_review": false, "location": {"page": 1}},
      {"key": "employee_name", "value": "Jordan Rivera", "needs_review": false, "location": {"page": 1}},
      {"key": "wages_box1", "value": 84500.00, "needs_review": false, "location": {"page": 1}},
      {"key": "federal_tax_withheld_box2", "value": 12180.00, "needs_review": false, "location": {"page": 1}},
      {"key": "social_security_wages_box3", "value": 84500.00, "needs_review": false, "location": {"page": 1}}
    ]
  }
}

初始化配置

open "https://www.deepread.tech/dashboard/?utm_source=clawhub"
export DEEPREAD_API_KEY="sk_live_your_key_here"

尚未获取 API 密钥？运行：clawhub install uday390/deepread-agent-setup —— agent 将通过 OAuth 设备授权流程自动为您获取密钥。

数据结构定义（W-2）

{
  "type": "object",
  "properties": {
    "form_type":   {"type": "string", "description": "税务表格类型，例如 W-2、1099-NEC、1040"},
    "tax_year":    {"type": "number", "description": "表格所载纳税年度"},
    "employer_name": {"type": "string", "description": "雇主 / 付款方名称"},
    "employer_ein":  {"type": "string", "description": "雇主 EIN"},
    "employee_name": {"type": "string", "description": "雇员 / 收款方姓名"},
    "employee_ssn_last4": {"type": ["string","null"], "description": "仅 SSN 最后四位数字"},
    "wages_box1":    {"type": "number", "description": "第 1 栏 — 工资、小费及其他报酬"},
    "federal_tax_withheld_box2": {"type": "number", "description": "第 2 栏 — 扣缴的联邦所得税"},
    "social_security_wages_box3": {"type": ["number","null"], "description": "第 3 栏 — 社会保障工资额"},
    "medicare_wages_box5": {"type": ["number","null"], "description": "第 5 栏 — 医疗保险工资额"},
    "state_wages_box16": {"type": ["number","null"], "description": "第 16 栏 — 州工资额"}
  }
}

对于 1099-NEC，请将字段替换为 payer_name、payer_tin、recipient_name、nonemployee_compensation_box1。模式保持一致——每项栏位均需明确描述，DeepRead 即可准确映射。

调用方式（cURL）

curl -s -X POST https://api.deepread.tech/v1/process 
  -H "X-API-Key: $DEEPREAD_API_KEY" 
  -F "file=@w2.pdf" 
  -F 'schema={"type":"object","properties":{"form_type":{"type":"string"},"tax_year":{"type":"number"},"employer_name":{"type":"string"},"wages_box1":{"type":"number"},"federal_tax_withheld_box2":{"type":"number"}}}'
# → {"id":"...","status":"queued"} — 随后使用 GET 请求查询 /v1/jobs/{id}

典型应用场景

• 报税软件 —— 从用户上传的 W-2/1099 文件中自动填充申报表，替代人工逐栏录入
• 信贷审批 / 风控建模 —— 基于税务文件验证申请人收入真实性
• 会计师事务所 —— 每年旺季批量导入客户税务表格
• 薪资审计 —— 将 W-2 汇总金额与内部薪资系统记录交叉核对

负责任地处理 PII —— 共享前务必脱敏

税务表格包含大量敏感信息，如 SSN、EIN 和地址。请仅提取必要字段（例如 employee_ssn_last4），并在向 deepread-pii 提交前对完整文档进行脱敏处理：

curl -X POST https://api.deepread.tech/v1/pii/redact -H "X-API-Key: $DEEPREAD_API_KEY" -F "file=@w2.pdf"

安装命令：clawhub install uday390/deepread-pii

使用建议

• 为每项栏位显式命名（例如“第 1 栏 — 工资…”），远优于仅用“wages”等模糊表述。
• 若需反复解析同一类表格？ 可在 https://www.deepread.tech/dashboard/optimizer 构建专属解析蓝图，提升准确率 20–30%。
• 关注 needs_review 字段 —— 仅该标志为 true 的栏位才需人工复核。

相关 DeepRead 技能

• deepread-ocr —— 通用文档识别提取能力 —— clawhub install uday390/deepread-ocr
• deepread-pay-stubs —— 从工资单中验证收入 —— clawhub install uday390/deepread-pay-stubs
• deepread-pii —— 敏感数据自动脱敏 —— clawhub install uday390/deepread-pii

支持资源

• 控制台：https://www.deepread.tech/dashboard
• 邮箱支持：support@deepread.tech

免费开始使用： https://www.deepread.tech/dashboard/?utm_source=clawhub