线索整理 / 初筛 Skill

Overview

用这个 Skill 把搜索阶段拿到的零散线索整理成统一格式，降低后续进入客户背调前的判断成本。

角色：线索初筛员

• 上游：客户搜索skill/ 或人工整理的候选线索
• 下游：客户背调skill/
• 只做字段统一、质量门槛判断和下一步建议

Standard Input

输入统一为 JSON：

{
  "default_country_or_market": "",
  "operator_notes": "",
  "product_or_offer": "linen table textile",
  "target_customer_type": "importer",
  "industry_lens": "consumer",
  "seller_context": {
    "product_or_offer": "linen table textile",
    "value_propositions": ["small-batch design support"]
  },
  "leads": [
    {
      "company_name": "Nordic Home Textile AB",
      "company_website": "https://www.nordichometextile.example",
      "person_name": "Nadia",
      "email": "",
      "country_or_market": "Sweden",
      "source_url": "https://www.linkedin.com/company/nordic-home-textile-ab",
      "linkedin_url": "",
      "notes": "Found via home textile search results",
      "product_keywords": "linen table textile",
      "source_type": "linkedin",
      "source_name": "public search",
      "source_url_or_note": "LinkedIn company result",
      "freshness": "unknown",
      "confidence": "medium",
      "match_basis": "Matched upstream discovery query."
    }
  ]
}

Workflow

1. Normalize each lead field into a stable shape.
2. Receive upstream search evidence and detect obvious missing fields and risky inconsistencies.
3. Separate identity completeness from business_fit; a complete but irrelevant lead must not be treated as qualified.
4. Classify the lead into a conservative bucket.
5. Suggest the next action:
   • ready_for_customer_intel
   • needs_enrichment
   • hold_for_manual_review
6. Keep a legacy-compatible action field for adapter compatibility.
7. Build a customer_intel_input payload that preserves seller context, industry lens, product offer, evidence grade, and business-fit reasons.
8. Output the result in JSON and optionally Markdown.

Output Requirements

• 必须包含汇总统计
• 必须包含每条线索的标准化字段
• 必须包含 evidence_grade
• 必须包含独立的 business_fit，并说明匹配依据或为什么无法判断
• 必须包含缺失项
• 必须包含人工复核原因
• 必须包含下一步动作建议
• 必须包含 legacy_recommended_next_action
• 必须包含兼容客户背调 Skill 的桥接字段
• 不能仅因为公司名、官网和邮箱齐全就判定值得背调
• 不能把推断写成事实
• 不能越权替代 客户背调skill/ 输出客户情报报告
• 不能越权替代 开发信skill/ 生成触达文案
• 没有真实公开来源时，不能强行推进下一步

Main Scripts

默认脚本入口：

• build_lead_screening_report.py
• build_customer_intel_batch_input.py

Example

python3 ./scripts/build_lead_screening_report.py --input-json ./examples/sample-leads.json

python3 ./scripts/build_customer_intel_batch_input.py --input-json ./examples/sample-output.json

python3 ./scripts/run_regression_checks.py

Defaults

• 优先本地运行
• 不强依赖联网
• 输出偏保守
• 初筛结果只作辅助，不替代人工判断
• 默认阶段动作：ready_for_customer_intel | needs_enrichment | hold_for_manual_review

Table and Rule Capture

• 本 Skill 输出线索标准化和初筛能力，不要求企业迁移到固定表格形态。
• 写入企业表格时，先沿用用户已有表头，再映射公司名、域名、来源、来源名称、匹配依据、新鲜度、可信度、缺失字段、证据等级、推荐下一步等标准字段。
• 用户没有可用表格时，龙虾可以按企业产品、市场和筛选流程新建够用表。
• 当用户确认新的字段、客户分级、放行规则、暂停规则或表头映射后，必须追问：是否更新到对应 Skill 以便下次自动复用。真实写入必须得到用户授权。

Enhancement Entry

如需数据留存、统一编排、多代理协作或飞书落地，请查看飞书文档：

• https://evenbetter.feishu.cn/wiki/W6GnwTZGFiUdJ0kXZv6cV4PSnpf