通用开发完全不需要，特别是在用 codex 的情况下，codex 的 agent 自带的提示词和工作流优于任何目前的通用 skill
但是公司内部流程是需要的，比如下面是一个平安保险 pdf 的查询 skill
```
---
name: pingan-health-pdf
description: Look up Ping An Health Insurance public disclosure PDF material URLs by product name and return JSON data. Use when a user asks for 平安健康险 / 平安健康保险 product PDFs such as 产品条款, 费率表, 产品说明书, 特定清单, 现金价值表, 合同条款, or public disclosure materials by product name.
---
# Ping An Health PDF Lookup
## Goal
Given a Ping An Health Insurance product name, find public disclosure material URLs from each matched product's `contentReference` JSON. Search both product pools by default:
- 在售保险产品: `12105675798371923550`
- 停售保险产品: `14411518807585587486`
Known material keys:
- `productTerm`: 产品条款
- `rate`: 产品费率表
- `instructions`: 产品说明书
- `specialList`: 产品特定清单
- `cashValue`: 产品现金价值表
Return JSON only. Do not summarize as prose unless the user explicitly asks for explanation.
## Workflow
1. Query both tags using the product name.
2. For each matched product, read `contentView.contentReference`.
3. Fetch the `contentReference` URL; it returns JSON containing PDF/material entries.
4. Normalize every non-empty material entry into `materials[]`.
5. Return exactly one JSON object with a top-level `data` field.
6. If no exact result is found, retry with a shorter title fragment. If still none, return `{"data":[]}`.
## Product Query
Endpoint:
```text
POST
https://mcore.health.pingan.com/bff/content/info/queryHistoryVersionList```
Headers:
```text
content-type: application/json
origin:
https://health.pingan.comreferer:
https://health.pingan.com/```
Body template:
```json
{
"pageSize": 10,
"pageNo": 1,
"tagId": "<TAG_ID>",
"type": 7,
"leaf": true,
"title": "<PRODUCT_NAME>"
}
```
Important response fields:
- `data.data[].productCode`
- `data.data[].onSaleAt`
- `data.data[].offSaleAt`
- `data.data[].productLevelCode`
- `data.data[].contentView.title`
- `data.data[].contentView.publishedAt`
- `data.data[].contentView.contentReference`
- `data.data[].contentView.contentContentTags[].tag.name`
## Output Format
Return exactly one valid JSON object:
```json
{
"data": [
{
"title": "产品名称",
"product_code": "产品代码",
"product_level_code": "产品分级",
"on_sale_at": "开售时间",
"off_sale_at": "停售时间",
"published_at": "发布时间",
"status": ["在售保险产品"],
"content_reference": "contentReference URL",
"materials": [
{
"key": "productTerm",
"label": "产品条款",
"filename": "文件名",
"url": "PDF URL",
"raw": {}
}
]
}
]
}
```
Output rules:
- Output JSON only.
- Do not wrap the JSON in Markdown fences.
- Do not add explanatory prose before or after the JSON.
- `data` is always an array.
- `materials` is always an array.
- Include every non-empty material key and every file entry.
- Use the Chinese label for known material keys; preserve unknown keys with `label` equal to the key.
- For object entries, set `filename` from `filename`, `name`, or the label. Set `url` from `url` or `link`, or empty string.
- For non-object entries, set `filename` to the label, `url` to the string value, and `raw` to the original value.
## Python Option
Use this when Python 3 is available. It only uses the Python standard library and prints the required JSON shape.
```bash
PRODUCT_NAME='平安互联网全家福住院医疗保险' python3 - <<'PY'
import json
import os
import urllib.request
```