Claude Code Skills 开发实战：从零造一个 Mock 数据生成 Skill

flowchart LR
    A["启动会话"] --> B["仅加载所有 Skill 的<br/>name + description"]
    B --> C{"用户输入<br/>是否匹配某 Skill？"}
    C -->|匹配| D["读取该 Skill 的 SKILL.md 正文"]
    C -->|不匹配| E["不加载，省 token"]
    D --> F{"正文是否引用<br/>脚本/参考文档？"}
    F -->|引用| G["按需读取 scripts/ references/"]
    F -->|不引用| H["直接执行"]
    G --> H

/skill-creator 帮我创建一个名为 mock-data 的 Skill，作用是根据用户描述的数据结构生成贴近真实的 Mock 测试数据

.claude/skills/mock-data/
├── SKILL.md          # 主指令文件（含 frontmatter）
├── references/        # 可选：详细参考文档
└── scripts/           # 可选：可执行脚本
    └── gen_mock.py

---
name: mock-data
description: Generate realistic mock/test data (Chinese names, IDs, phone numbers, addresses, nested JSON, arrays) from a data structure description. Use when the user needs fake sample data, test fixtures, seed data, dummy JSON for API development, or asks to "造点测试数据/生成假数据/mock 一下". Outputs deterministic, constraint-respecting data.
---

# Mock Data Generation

You generate realistic mock data that respects the user's stated constraints.
Do NOT hallucinate fields the user didn't ask for. When a field type is
ambiguous, ask one clarifying question, then proceed.

## Workflow

1. Parse the user's data structure description into a field list.
2. For each field, pick the most realistic generator (see references/field-types.md).
3. Respect constraints: count, uniqueness, value ranges, locale, format.
4. Output as the requested format (JSON / CSV / SQL INSERT / Python dict).
5. If count > 50, stream the output to a file instead of printing inline.

## Realism rules

- Chinese names: real-style surnames + given names, not "张三"/"李四" repeats.
- Phone numbers: start with valid Chinese prefixes (13x/15x/17x/18x/19x).
- ID numbers: valid checksum (use scripts/gen_mock.py --id to compute).
- Dates: spread across a plausible range, not all "2026-01-01".
- Money: realistic magnitude per domain (salary 5k–50k, not 999999).

## When to stop and ask

- If the user requests a field you cannot map to any generator.
- If two constraints conflict (e.g., "age between 0-3 AND employed").
- If output would exceed 200 rows — confirm the file path first.

#!/usr/bin/env python3
"""Mock data generator. Zero external deps.
Usage: gen_mock.py --field id --count 5
       gen_mock.py --field phone --count 3
"""
import argparse, json, random, sys

SURNAME = list("王李张刘陈杨黄赵吴周徐孙马朱胡郭何高林罗")
GIVEN = ["伟","芳","娜","秀英","敏","静","丽","强","磊","洋",
         "艳","勇","军","杰","涛","明","超","秀兰","霞","平"]
PHONE_PREFIX = ["133","135","137","150","159","170","180","188","199"]

def gen_id():
    # Simplified 18-digit ID with valid-ish region + checksum
    region = random.choice(["110101","310115","440305","510107"])
    birth = f"{random.randint(1960,2008)}{random.randint(1,12):02d}{random.randint(1,28):02d}"
    seq = f"{random.randint(0,999):03d}"
    base = region + birth + seq
    # Weighted checksum (GB 11643-1999)
    w = [7,9,10,5,8,4,2,1,6,3,7,9,10,5,8,4,2]
    code = "10X98765432"
    s = sum(int(d)*wi for d, wi in zip(base, w))
    return base + code[s % 11]

def gen_name():
    return random.choice(SURNAME) + "".join(random.sample(GIVEN, random.randint(1,2)))

def gen_phone():
    return random.choice(PHONE_PREFIX) + "".join(str(random.randint(0,9)) for _ in range(8))

GENERATORS = {"id": gen_id, "name": gen_name, "phone": gen_phone}

if __name__ == "__main__":
    p = argparse.ArgumentParser()
    p.add_argument("--field", required=True)
    p.add_argument("--count", type=int, default=1)
    args = p.parse_args()
    fn = GENERATORS.get(args.field)
    if not fn:
        sys.exit(f"unknown field: {args.field}")
    print(json.dumps([fn() for _ in range(args.count)], ensure_ascii=False))

## Using the bundled script

For ID numbers, phone numbers, and names, prefer calling the bundled script
to guarantee format validity:

    python3 scripts/gen_mock.py --field id --count 10
    python3 scripts/gen_mock.py --field phone --count 5

Compose the script's output into the larger structure the user asked for.
Do NOT hand-compute ID checksums — always delegate to the script.

# references/field-types.md

| 字段语义 | 生成策略 | 示例 |
|---------|---------|------|
| email | 拼音名+域名，域名随机自 gmail/qq/163 | `[email protected]` |
| address | 省+市+区+随机街道门牌 | `广东省深圳市南山区科技园路1号` |
| company | 行业词+公司类型 | `星河科技有限公司` |
| price | 按商品类目量级 | 电子产品 99–9999，生鲜 1–200 |
| timestamp | 最近 N 天内随机 ISO8601 | `2026-06-14T03:21:00Z` |
| boolean | 0.3–0.7 比例 True | `true` |
| enum | 从用户给的候选集随机 | `["active","pending","closed"]` |

For the full field-type mapping table, consult references/field-types.md
when you encounter a field type not covered in the realism rules above.

---
name: mock-data
description: ...
allowed-tools: Read, Write, Bash(python3 scripts/gen_mock.py:*)
---

.claude/skills/mock-data/
├── SKILL.md
├── references/
│   └── field-types.md
└── scripts/
    └── gen_mock.py

sequenceDiagram
    participant U as 用户
    participant C as Claude Code
    participant S as mock-data Skill
    participant P as gen_mock.py
    U->>C: 一句话需求
    C->>S: 命中 mock-data 触发词<br/>加载 SKILL.md
    S->>C: 返回 workflow + 调用规则
    C->>P: Bash: gen_mock.py --field phone --count 1000
    P-->>C: 返回合规手机号数组
    C->>P: Bash: gen_mock.py --field id --count 1000
    P-->>C: 返回合规身份证数组
    C->>C: 组装商品/地址/时间/状态<br/>(查 references/field-types.md)
    C->>U: 写入 orders.json (1000 行)
    C->>U: 汇总：字段、约束、行数

[
  {"order_id":"ORD-20260710-0001","buyer":"陈艳霞","phone":"137xxxx2841",
   "address":"广东省深圳市南山区科苑南路8号","item":"无线蓝牙耳机",
   "unit_price":199,"qty":2,"ts":"2026-06-22T11:03:00Z","status":"paid"},
  {"order_id":"ORD-20260710-0002","buyer":"林涛","phone":"188xxxx7710",
   "address":"北京市海淀区中关村大街1号","item":"机械键盘",
   "unit_price":459,"qty":1,"ts":"2026-07-01T08:45:00Z","status":"shipped"}
]