Tongyong Shenhe

通用内容审核 Skill。配置驱动，适用于所有 d.php 后台站点。内置审核规则自动判断 + 可选技术部API增强。其他组只需填写站点账号密码即可使用，审核规则可自行修改适配。

Security Scan

The skill mostly does what it says (automated moderation of d.php backends) but there are mismatches and privacy/exfiltration risks you should understand before using it (notably requiring the TOTP seed, relying on curl via VPN, and an optional configurable external moderation API).

Purpose & Capability

The name/description (generic content-moderation for d.php sites) aligns with the included code (review.py) and rules.json: the script logs into an admin panel, fetches pending items and submits review decisions. However the registry metadata lists no required binaries while the code uses the system curl binary (via subprocess). The need for admin username/password is expected; requiring the TOTP seed (not just a one-time code) is more sensitive but explainable for unattended automation.

Instruction Scope

SKILL.md and DEPLOY/USAGE instruct connecting a VPN, providing admin credentials and the TOTP seed, and optionally sending content to a '技术部' moderation API. The docs also explicitly suggest sending rules.json to an external AI (Claude) to edit rules — that directs you to transmit configuration/content externally. The script will POST content it extracts to any api_url you configure, so if you set a third‑party api_url the skill will send item content (potentially sensitive) offsite. These behaviors broaden scope beyond local-only moderation and raise data-leak risk.

Install Mechanism

No install spec (instruction-only plus a Python script) — low install risk. It depends on Python stdlib and optionally requests; DEPLOY asks to pip install requests only if API is used. The code executes curl via subprocess; this is not an installation-time download but runtime use of a system binary. No remote archives or opaque installers are fetched by the skill itself.

Credentials

The skill does not request environment variables, but it requires sensitive credentials in a local config file: admin username/password and the TOTP seed. Requiring the TOTP seed (a persistent secret that can recreate 2FA tokens) is high-risk — many teams would avoid giving out seeds and prefer device-bound or short-lived approaches. The optional moderation API requires api_url and api_key; because api_url is arbitrary, a configured external service could receive all moderated content and any metadata added to requests (exfiltration risk). The config.example sets a non-obvious default api_url (https://zyaokkmo.cc) — this should be verified before use.

Persistence & Privilege

The skill is not always-enabled and doesn't request system-wide privileges. It creates a temporary cookie file for sessions and cleans up; it does not modify other skills or global agent configuration. Autonomous invocation is allowed (platform default) but not a new privilege introduced by this skill.

What to consider before installing

Before installing or running this skill, consider the following: 1) The script requires admin credentials and the TOTP seed — giving the seed grants long-term 2FA capability, so avoid supplying it unless you trust the operator; prefer a service account with limited permissions or use manual/interactive TOTP entry. 2) The skill uses system curl and expects a VPN interface (ppp0); the registry metadata did not declare curl — verify your environment and run in an isolated/test account first. 3) The moderation API is optional but accepts an arbitrary api_url and api_key; double-check the URL (the example domain looks unfamiliar). If you configure an external API, you will be sending full item text and metadata offsite — only enable this for trusted internal endpoints. 4) The docs explicitly suggest sending rules.json (and possibly example content) to external AI services (e.g., Claude) — this can leak policy or sample content; avoid sending sensitive examples. 5) Run the tool in --dry-run mode first; audit review.py yourself (search for unexpected network endpoints or hidden behavior), and consider executing it from a network-isolated environment or with network controls to prevent unintended exfiltration. If you want, I can list exact lines in review.py to inspect and suggest safer configuration alternatives (e.g., avoid storing TOTP seed, restrict api_url to internal hostnames).

通用内容审核 Skill

概述

配置驱动的通用内容审核技能，适用于公司所有基于 d.php 框架的后台站点。内置通用审核规则（联系方式检测、违禁内容、广告导流等），开箱即用。各组可根据自身站点需求修改 rules.json 适配。

核心特性：

• 开箱即用：填入账号密码即可运行，内置审核规则覆盖常见违规类型
• 规则可定制：rules.json 结构清晰，可直接让 AI 帮忙修改适配
• 可选 API 增强：配置技术部审核 API key 后可获得 AI 模型双重审核
• 安全机制：dry-run 模式先验证再上线

工作流

VPN 连接（ppp0 接口）
  │
  ├─ TOTP 登录后台
  │
  ├─ 拉取待审列表
  │
  ├─ 逐条审核判断
  │    │
  │    ├─ 第1层：本地规则（rules.json）
  │    │    ├─ 命中 → 自动拒绝 + 原因
  │    │    └─ 未命中 → 进入第2层
  │    │
  │    └─ 第2层：技术部 API（可选，需配置 api_key）
  │         ├─ rejected → 自动拒绝
  │         ├─ flagged → 不提交，留给人工
  │         └─ approved → 通过
  │
  │    如果未配置 API：本地规则通过即自动通过
  │
  └─ 输出统计

内置审核规则

rules.json 预置以下规则（从现有三个站点提取的通用标准）：

规则	默认启用	说明
联系方式检测	是	微信/QQ/TG/手机号/连续数字
网址链接检测	是	URL、域名、网址
广告导流话术	是	"加我"、"私聊"、"扫码"等
未成年人内容	是	"学生妹"、"初中"、"幼女"等
违规服务描述	是	色情服务关键词
诈骗赌博黑产	是	赌博、刷单、彩票等
多地区限制	否	"全国"、"可空降多地"（本地服务站点可启用）
联系方式格式	否	phone 字段格式校验（有 phone 字段的站点启用）
地址详细度	否	地址质量检查（有 address 字段的站点启用）
价格范围	否	价格区间+整百校验（有 price 字段的站点启用）
标题格式	否	标题格式校验（VIP 资源类站点启用）

默认关闭的规则适用于特定业务场景，启用只需在 rules.json 中将 "enabled": false 改为 true。

适配原理

公司所有站点后台基于同一套 d.php 框架，接口模式固定：

操作	路径风格（默认）	查询参数风格
登录	/d.php/admin/login/doLogin	d.php?mod=login&code=dologin
拉取列表	/d.php/admin/{模块}/listAjax	d.php?mod={模块}&code=listAjax
提交结果	/d.php/admin/{模块}/verifyStatus	d.php?mod={模块}&code=verifyStatus

模块清单

文件	功能
review.py	主审核脚本
rules.json	审核规则（可自定义）
config.example.json	配置模板
SKILL.md	本文档
DEPLOY.md	部署指南
USAGE.md	使用指南

配置文件

配置项	必填	说明
site.base_url	是	后台地址
site.module	是	审核模块名
auth.username	是	登录用户名
auth.password	是	登录密码
auth.totp_seed	是	TOTP 密钥
moderation.content_fields	否	送审字段（默认 ["title", "content"]）
moderation.api_key	否	技术部 API 密钥（不填则仅用本地规则）

输出示例

[14:30:01] ═══ 通用审核 [茶馆大厅] (LIVE) ═══
[14:30:01] 已加载 6 条审核规则
[14:30:01] 正在登录...
[14:30:02] 登录成功
[14:30:02] 待审总量: 45 条
[14:30:02] ── 第 1 页（45 条）──
[14:30:02]   [1001] PASS
[14:30:02]   [1002] REJECT [contact_keywords]: 含联系方式或疑似导流信息，审核失败
[14:30:02]   [1003] PASS
[14:30:02]   [1004] REJECT [underage_content]: 平台禁止发布未成年人相关信息，审核失败
...
[14:30:15] 完成: 共45条 | 通过32 | 拒绝8 | 待复审0 | 跳过3 | 异常2