门店客流分析

门店客流与转化漏斗分析工具。基于AIoT客户行为数据（customerFunnel + behaviorFunnel）。 核心能力： 1. 双漏斗结合分析（customerFunnel客户分层 + behaviorFunnel试用行为） 2. 五步分析法（获取数据→解析customerFunnel→解析behav...

Security Scan

The skill implements the stated retail funnel analysis, but it imports a local api_client via a hard-coded user path and relies on an external data-fetch function (which likely needs credentials) without declaring or documenting those credentials — this mismatch could cause unexpected access to local modules or secrets.

Purpose & Capability

The code and SKILL.md match the described purpose (fetch customerFunnel and behaviorFunnel, compute conversion metrics). However, the implementation inserts a hard-coded sys.path to '/Users/yangguangwei/.openclaw/workspace-front-door' to import api_client, which is a user-specific path and unusual for a distributable skill; this is a coherence / portability oddity (works only in that environment) and could cause the skill to pick up arbitrary local code.

Instruction Scope

SKILL.md instructs the agent to fetch data from an API endpoint and run the analysis — consistent with the code. The code delegates all network/data access to api_client.get_copilot_data('/api/v1/store/dashboard/bi?...'). The SKILL.md does not specify how api_client is configured or what credentials it uses. Because data fetching is outsourced to api_client, the runtime behavior depends entirely on that module (which may perform network calls, use local credentials, or contact unexpected endpoints).

Install Mechanism

No install spec or external downloads are present; the skill is instruction+code only. Nothing is written to disk by an installer in the provided files.

Credentials

The skill declares no required environment variables or credentials but imports a local api_client that almost certainly requires configuration (API host, auth tokens). The hard-coded insertion of a local absolute path means the skill can import code from a user's filesystem location, which could access local secrets or system-configured credentials. The lack of declared credential requirements is a mismatch and a potential avenue for unexpected access to sensitive data.

Persistence & Privilege

The skill does not request always:true, does not modify other skills or global agent settings, and returns results without persisting changes. No elevated persistence behavior observed.

What to consider before installing

This skill appears to implement the documented analysis but has two issues you should verify before installing: 1) Hard-coded import path: analyze.py inserts '/Users/yangguangwei/.openclaw/workspace-front-door' into sys.path and imports api_client. That means the skill will load code from that local path if present. Verify who authored the api_client at that location and what it does. If you don't control or trust that path, the skill could execute arbitrary local code. 2) Undeclared credentials/config: The skill calls get_copilot_data(...) to fetch store data but doesn't declare required API host or authentication environment variables. Inspect api_client.get_copilot_data to see where it sends requests and which credentials it uses. Ensure it doesn't send data to unexpected endpoints or read secrets from your environment (e.g., ~/.aws, token files, or other local config). Recommended actions before use: - Open and review the api_client implementation that will be imported in your environment (or run the skill in an isolated sandbox where you control api_client). - Replace the hard-coded sys.path insertion with a documented import/install mechanism (or vendor a minimal, audited api client in the skill) and require explicit env vars for API host and token. - Run the skill in a restricted environment or with network monitoring to confirm it only calls the intended API endpoint and does not exfiltrate data. If you can provide the api_client source or confirm where get_copilot_data sends requests and how it's authenticated, I can reassess and raise or lower the concern level.

客流分析 Skill

技能名称

traffic-analysis

功能描述

基于 AIoT 客户行为数据（customerFunnel + behaviorFunnel），分析门店客流和转化漏斗，诊断客流问题和转化效率问题。

核心特点

1. 双漏斗结合分析

• customerFunnel: 客户分层（人的数量）
• behaviorFunnel: 试用行为（行为次数）
• 结合分析: 人 + 行为 = 完整的客户行为画像

2. 五步分析过程

1. 获取数据 — 从 API 拉取 customerFunnel 和 behaviorFunnel
2. 解析 customerFunnel — 客户分层（互斥桶验证）
3. 解析 behaviorFunnel — 试用行为（数据关系验证）
4. 计算转化率 — 客户分层转化 + 行为转化效率
5. 综合诊断 — 识别核心问题

3. 完整逻辑标注

每个转化率指标都包含：

• 逻辑说明 — 这个指标代表什么
• 计算公式 — 具体怎么算
• 原因解释 — 为什么这样算
• 计算过程 — 展示具体数字和步骤

核心逻辑

互斥桶模型（customerFunnel）

有效客户（100%，互斥）
├── 普通客户（无试用）
├── 潜在客户（普通试用，无深度）
├── 意向客户（深度试用，无成交）
└── 成交客户（有成交）

互斥原则：每个客户只计入最深层的桶 验证：普通 + 潜在 + 意向 + 成交 = 有效客户

行为分层（behaviorFunnel）

总试用次数
├── 普通试用次数（交互强度弱）
└── 深度试用次数（交互强度高）

关系：普通试用 + 深度试用 = 总试用次数

转化漏斗

1. 客户分层转化率（基于互斥桶）

a) 潜在→意向转化率

• 逻辑: 从"有普通试用"到"有深度试用"的转化
• 公式: (意向客户 + 成交客户) / (潜在客户 + 意向客户 + 成交客户)
• 原因: 意向客户和成交客户都有深度试用，成交客户是从意向转化来的
• 示例:

  上期: (65+296)/(19+65+296) = 361/380 = 95.0%
  本期: (78+188)/(27+78+188) = 266/293 = 90.8%
  

b) 意向→成交转化率

• 逻辑: 从"有深度试用"到"成交"的转化
• 公式: 成交客户 / (意向客户 + 成交客户)
• 原因: 只有意向客户和成交客户有深度试用，成交是意向的下一步
• 示例:

  上期: 296/(65+296) = 296/361 = 82.0%
  本期: 188/(78+188) = 188/266 = 70.7%
  

2. 行为转化效率（基于 behaviorFunnel）

a) 试用→成交转化率

• 逻辑: 总试用次数中有多少转化为成交
• 公式: 成交件数 / 总试用次数
• 示例:

  上期: 341/7121 = 4.8%
  本期: 201/4388 = 4.6%
  

b) 深度试用→成交转化率

• 逻辑: 深度试用次数中有多少转化为成交
• 公式: 成交件数 / 深度试用次数
• 示例:

  上期: 341/927 = 36.8%
  本期: 201/555 = 36.2%
  

3. 人+行为综合指标

a) 高价值客户人均深度试用

• 逻辑: 有深度试用的客户，平均每人深度试用多少次
• 公式: 深度试用次数 / (意向客户 + 成交客户)
• 原因: 意向客户和成交客户都产生了深度试用行为
• 示例:

  上期: 927/(65+296) = 927/361 = 2.6次/人
  本期: 555/(78+188) = 555/266 = 2.1次/人
  

b) 高价值客户人均成交件数

• 逻辑: 有深度试用的客户，平均每人成交多少件
• 公式: 成交件数 / (意向客户 + 成交客户)
• 示例:

  上期: 341/361 = 0.94件/人
  本期: 201/266 = 0.76件/人
  

使用方法

import sys
sys.path.insert(0, '~/.openclaw/skills/traffic-analysis')
from analyze import analyze

# 运行完整分析
result = analyze(
    store_id="416759_1714379448487",
    from_date="2026-03-01",
    to_date="2026-03-25",
    store_name="正义路60号店"
)

# 输出包含:
# - 完整分析过程（打印到控制台）
# - 结构化结果（返回值）

输出示例

控制台输出（完整分析过程）

======================================================================
客流分析报告 - 正义路60号店
分析周期: 2026-03-01 至 2026-03-25
======================================================================

【第一步：获取数据】
API: /api/v1/store/dashboard/bi
参数: storeId=416759_1714379448487, fromDate=2026-03-01, toDate=2026-03-25

【第二步：解析 customerFunnel（人的数量）】
...

【第三步：解析 behaviorFunnel（行为次数）】
...

【第四步：计算转化率（核心分析）】
...

【第五步：综合诊断】
转化率变化汇总:
  潜在→意向:     95.0% → 90.8% (-4.2%)
  意向→成交:     82.0% → 70.7% (-11.3%)
  ...

核心问题识别:
  1. 🔴 客流显著下滑 (有效客户-14.6%)
  2. 🔴 意向→成交转化率显著下降 (-11.3%)
  ...

结构化结果（返回值）

{
  "status": "ok",
  "funnel": {
    "customer_groups": {"name": "有效客户", "current": 350, "previous": 410, "change_pct": -14.6},
    "normal_groups": {"name": "普通客户", "current": 57, "previous": 30, "change_pct": 90.0},
    "potential_groups": {"name": "潜在客户", "current": 27, "previous": 19, "change_pct": 42.1},
    "intent_groups": {"name": "意向客户", "current": 78, "previous": 65, "change_pct": 20.0},
    "deal_groups": {"name": "成交客户", "current": 188, "previous": 296, "change_pct": -36.5}
  },
  "conversion_rates": {
    "潜在到意向": {"上期": 0.95, "本期": 0.908, "变化": -0.042},
    "意向到成交": {"上期": 0.82, "本期": 0.707, "变化": -0.113},
    "试用到成交": {"上期": 0.048, "本期": 0.046, "变化": -0.002},
    "深度到成交": {"上期": 0.368, "本期": 0.362, "变化": -0.006}
  },
  "per_customer_metrics": {
    "人均深度试用": {"上期": 2.6, "本期": 2.1},
    "人均成交件数": {"上期": 0.94, "本期": 0.76}
  }
}

诊断标准

问题	指标	阈值	说明
客流下滑	有效客户变化	<-10%	进店人数显著减少
转化效率下降	意向→成交转化率变化	<-5pp	深度客户不成交
试用深度不足	潜在→意向转化率变化	<-5pp	客户试得浅了
人均试用下降	人均深度试用变化	<-0.3次	高价值客户试得少了
人均购买下降	人均成交件数变化	<-0.1件	买得少了

依赖

• api_client.get_copilot_data() — 数据获取
• customerFunnel — 客户分层数据（互斥桶）
• behaviorFunnel — 试用行为数据

版本

v1.1.0 — 完整分析过程展示（五步分析法 + 逻辑标注 + 人+行为结合）