🛡️ LingxiGuard 灵犀盾 — AI Comment Spam Filter for WordPress / AI 反垃圾评论插件

AI-powered WordPress comment spam filter. Automatically detects spam and harmful comments by sending both the comment and the article content to an OpenAI-compatible Chat Completions API (DeepSeek by default), then moves spam straight to the trash. 装上就能用的 AI 反垃圾评论插件：把「评论 + 文章内容」交给 AI 判断，垃圾评论自动丢进回收站。

Keywords: WordPress plugin, AI comment spam filter, anti-spam, comment moderation, DeepSeek, OpenAI-compatible API, 评论过滤, 反垃圾评论, 评论审核

Table of Contents / 目录

• ✨ Features
• 🚀 Installation
• ⚙️ Configuration
• 🔌 API Compatibility
• 🧠 How It Works
• 📊 Screenshots
• 📦 Development / Build
• 📝 Changelog
• 📄 License

✨ Features / 功能特性

English:

• 🗑️ Trash instead of delete — spam comments go to the WordPress Trash, recoverable anytime
• 🧠 Context-aware judgment — sends the article title + content together with the comment, so AI judges in context, not just keywords
• 🔌 Any OpenAI-compatible API — configurable API base URL (default https://api.deepseek.com) and any model name (deepseek-v4-flash, deepseek-v4-pro, gpt-4o-mini, ...)
• 🧪 One-click test — verify your API key works with a sample spam comment before going live
• 💾 Smart caching — identical comments are judged only once per hour (saves tokens)
• 🛡️ Fail-open by default — if the API errors, comments pass through instead of being wrongly deleted (switchable)
• 👤 Skip logged-in users — admins and registered users are never blocked
• 📋 Audit log — last 100 decisions with AI reasoning, right on the settings page
• 🌍 Bilingual UI — Chinese-first interface with English descriptions

中文：

• 🗑️ 进回收站不直接删：垃圾评论放入回收站，误判可随时恢复
• 🧠 结合文章上下文判断：把文章标题和内容一起发给 AI，判断更准，不是简单关键词匹配
• 🔌 兼容任意 OpenAI 格式 API：API 地址可配置（默认 DeepSeek），模型名随意填
• 🧪 一键测试：上线前用示例垃圾评论验证 API 是否正常工作
• 💾 智能缓存：相同评论 1 小时内只调用一次 AI，省 token
• 🛡️ 失败默认放行：API 故障时评论正常通过，不会误删（可切换为严格模式）
• 👤 跳过已登录用户：管理员和注册用户评论不经过 AI
• 📋 审核日志：设置页直接查看最近 100 条 AI 判断及理由
• 🎛️ 极简设置页：装上就能用，高级选项全部折叠

🚀 Installation / 安装

English:

1. Download the latest release zip, or clone this repo into wp-content/plugins/lingxiguard/
2. Activate the plugin in Plugins → Installed Plugins
3. Go to Settings → AI Comment Filter
4. Paste your API key (create one at platform.deepseek.com), the API base URL and model are pre-filled with DeepSeek defaults
5. Click 🧪 One-click test — done!

中文：

1. 下载最新版 zip，或把本仓库克隆到 wp-content/plugins/lingxiguard/
2. 在后台「插件 → 已安装的插件」中启用
3. 进入「设置 → AI 评论过滤」
4. 填入 API Key（在 platform.deepseek.com 创建），API 地址和模型已默认填好 DeepSeek 配置
5. 点「🧪 一键测试」验证通过即可，完事！

⚙️ Configuration / 配置

English:

中文：

🔌 API Compatibility / API 兼容

English:

Uses the standard OpenAI Chat Completions protocol:

POST {api_base}/chat/completions
Authorization: Bearer {api_key}
Content-Type: application/json

{
  "model": "deepseek-v4-flash",
  "messages": [...],
  "response_format": { "type": "json_object" },
  "thinking": { "type": "enabled" },          // when thinking mode is on
  "reasoning_effort": "high"                  // when thinking mode is on
}

Works with DeepSeek, OpenAI, and any self-hosted / relay gateway that implements the same protocol. The AI must reply with a strict JSON object: {"spam": true|false, "category": "spam"|"harmful"|"normal", "reason": "..."}.

中文：

采用标准 OpenAI Chat Completions 协议（见上方示例）。兼容 DeepSeek、OpenAI 以及任何实现该协议的自建/中转服务。AI 必须返回严格 JSON：{"spam": true|false, "category": "spam"|"harmful"|"normal", "reason": "..."}。

🧠 How It Works / 工作原理

English:

1. A visitor submits a comment
2. WordPress runs its built-in checks first (Disallowed Comment Keys → spam, Moderation Keys → pending)
3. Only comments WordPress actually approves reach the plugin's pre_comment_approved hook — blocked ones skip the AI entirely, saving tokens
4. Article title + content (truncated) + comment + author info are sent to the AI
5. AI returns spam: true/false
6. Spam → stored directly as trash (recoverable); normal → proceeds through the original moderation flow
7. Result cached for 1 hour; API errors fail open (configurable)

中文：

1. 访客提交评论
2. WordPress 先跑内置检查（评论黑名单关键词 → 垃圾评论，审核关键词 → 待审）
3. 只有 WordPress 放行的评论才会进入插件的 pre_comment_approved 钩子——已被拦下的评论不再调 AI，省 token
4. 把「文章标题 + 内容（截断）+ 评论 + 评论者信息」发给 AI
5. AI 返回 spam: true/false
6. 垃圾 → 直接以 trash（回收站） 状态入库（可恢复）；正常 → 走原有审核流程
7. 结果缓存 1 小时；API 出错默认放行（可配置）

📊 Screenshots / 界面预览

📦 Development / Build / 开发与打包

English: The plugin is a single PHP file — no build step required. To package a release zip:

zip -r lingxiguard.zip lingxiguard/

中文： 插件只有一个 PHP 文件，无需构建。打包发布 zip 如上。

📝 Changelog / 更新日志

See CHANGELOG.md for the full history. 完整更新日志见 CHANGELOG.md。

📄 License / 许可证

GPL-2.0-or-later — free software. See LICENSE.

Star ⭐ if you find this useful — it helps the project grow! 觉得好用点个 Star 支持一下！