seo-mcp-server/dist/tools/aiContentDetect.js

SEO MCP Server (minimal) - ai_content_detect only

"use strict";
// aiContentDetect.ts
// 工具名称: ai_content_detect
// 位置: 内容生成之后、SEO质检(validate_seo_content)之前的硬性前置检查
// 作用: 调用 5118 AI内容检测API，对内容进行AI痕迹评估，输出 percent、逐段 linesscore、mintxt 等；
//      以阈值规则生成 pass/review/fail 决策、下一步动作(next_actions)与人类可读建议(suggestions)。
// API文档要点:
//  - 请求: POST http://apis.5118.com/aidetect
//  - Header: Authorization: <API_KEY>
//            Content-Type: application/x-www-form-urlencoded; charset=UTF-8
//  - Body: content=<string>  (100~6000字)
//  - 返回(官方示例):
//    {
//      "errcode": "0",
//      "errmsg": "",
//      "data": {
//        "linesscore": [ { "score": 2, "txt": "..." }, { "score": 3, "txt": "..." } ... ],
//        "percent": 43,
//        "mintxt": 0
//      }
//    }
// 字段含义与解读:
//  - errcode: "0" 表示成功；非"0"为错误
//  - errmsg: 错误说明
//  - data.mintxt: 0=长度合规(>=100字)，1=内容太短(硬性不通过)
//  - data.percent: AI占比(值越高AI痕迹越重)
//  - data.linesscore[].score: 段落分值; <3 为疑似AI段落
// 判定阈值(可配置):
//  - maxPercent(默认35): percent<=35 通过
//  - reviewPercent(默认55): 35<percent<=55 复核
//  - minLineScore(默认3): 任一段<3 触发复写建议(通常进入review)
// 决策输出:
//  - pass: boolean
//  - decision: "pass" | "review" | "fail"
//    * pass: mintxt==0 且 percent<=maxPercent 且 所有段score>=minLineScore
//    * review: (maxPercent<percent<=reviewPercent) 或 存在少量低分段
//    * fail: percent>reviewPercent 或 mintxt!=0 或 errcode!=0
// 建议输出(next_actions):
//  - "revise_intro_reduce_cliches": 引言去套话
//  - "add_verifiable_details": 补充可核验事实(时间/地点/数据来源)
//  - "split_long_sentences": 拆分长句
//  - "remove_template_phrases": 去模板化表达
//  - "reduce_keyword_bolding": 减少关键词加粗导致统计异常
//
// 注意: 接口存在“非严格JSON”返回的情况。本工具内置“宽松解析”: 先尝试JSON.parse；失败则进行预清洗与正则回退提取。
// 诊断输出(diagnostics):
//  - raw: 原始响应文本
//  - parser: "strict" | "relaxed"
//  - notes: 解析备注
Object.defineProperty(exports, "__esModule", { value: true });
exports.aiContentDetect = aiContentDetect;
// 预清洗: 针对缺逗号/引号粘连的简单修复(尽力而为，失败则回退正则提取)
function tryRelaxedParse(raw, notes) {
    try {
        return JSON.parse(raw);
    }
    catch {
        // 尝试移除前缀/后缀的非JSON字符
        let t = raw.trim();
        // 去掉可能出现的前缀 "Output:"、"> y" 等
        t = t.replace(/^[^{\[]+/, "");
        // 简单修正常见缺逗号情形: }{ -> },{   "0""errmsg" -> "0","errmsg"
        t = t.replace(/}\s*{/g, "},{");
        t = t.replace(/"0""/g, '"0","');
        t = t.replace(/]\s*"(percent|mintxt)"/g, '],"$1"');
        // 再尝试解析
        try {
            const obj = JSON.parse(t);
            notes.push("relaxed parse applied");
            return obj;
        }
        catch {
            notes.push("relaxed parse failed");
            return null;
        }
    }
}
// 正则回退提取: 仅抽取核心字段
function fallbackExtract(raw, notes) {
    const errcode = (raw.match(/"errcode"\s*:\s*"?(?<v>\d+)"?/)?.groups?.v) ?? "";
    const errmsg = (raw.match(/"errmsg"\s*:\s*"(?<v>[^"]*)"/)?.groups?.v) ?? "";
    const percentStr = (raw.match(/"percent"\s*:\s*(?<v>\d+)/)?.groups?.v) ?? "0";
    const mintxtStr = (raw.match(/"mintxt"\s*:\s*(?<v>\d+)/)?.groups?.v) ?? "1";
    // 行分数粗提取(注意该接口会把每段当作对象输出)
    const lineMatches = [...raw.matchAll(/{"score"\s*:\s*(\d+)\s*"txt"\s*:\s*"([^"]*)"}|{"score":\s*(\d+),"txt":"([^"]*)"}/g)];
    const linesscore = [];
    let idx = 1;
    for (const m of lineMatches) {
        const s = Number(m[1] || m[3] || 0);
        const txt = (m[2] || m[4] || "").trim();
        if (txt)
            linesscore.push({ idx: idx++, score: s, txt });
    }
    notes.push("fallback extractor used");
    return {
        errcode: errcode || "0",
        errmsg,
        percent: Number(percentStr),
        mintxt: Number(mintxtStr),
        linesscore,
    };
}
async function aiContentDetect(input) {
    const { content, thresholds } = input;
    const apiKey = globalThis.__SEO_MCP_AI_DETECT_API_KEY__;
    if (!apiKey) {
        throw new Error('AI-Detect apiKey is not configured. Please set env SEO_MCP_AI_DETECT_API_KEY and restart seo-mcp.');
    }
    const maxPercent = thresholds?.maxPercent ?? 35;
    const reviewPercent = thresholds?.reviewPercent ?? 55;
    const minLineScore = thresholds?.minLineScore ?? 3;
    // 基本校验
    const len = (content || "").length;
    if (len < 100 || len > 6000) {
        return {
            errcode: "length_error",
            errmsg: `content length ${len} not in [100,6000]`,
            mintxt: len < 100 ? 1 : 0,
            percent: 0,
            linesscore: [],
            pass: false,
            decision: "fail",
            next_actions: ["increase_content_length_min_100"],
            suggestions: ["补充细节与事实，确保正文长度≥100字"],
            diagnostics: { parser: "strict", raw: "", notes: [] },
        };
    }
    // 发起请求
    const qs = new URLSearchParams();
    qs.set("content", content);
    const resp = await fetch("http://apis.5118.com/aidetect", {
        method: "POST",
        headers: {
            Authorization: apiKey,
            "Content-Type": "application/x-www-form-urlencoded; charset=UTF-8",
            Accept: "application/json",
        },
        body: qs.toString(),
    }).catch((e) => ({
        ok: false,
        status: 0,
        text: async () => String(e),
    }));
    const raw = await resp.text?.() ?? "";
    const notes = [];
    let parser = "strict";
    // 解析响应
    let data = null;
    try {
        data = JSON.parse(raw);
    }
    catch {
        const obj = tryRelaxedParse(raw, notes);
        if (obj) {
            data = obj;
            parser = "relaxed";
        }
    }
    let errcode = "0";
    let errmsg = "";
    let percent = 0;
    let mintxt = 1;
    let linesscore = [];
    if (data && typeof data === "object") {
        // 兼容 data 包裹
        const container = data.data ? data : { data };
        errcode = String(data.errcode ?? "0");
        errmsg = String(data.errmsg ?? "");
        const d = container.data || {};
        percent = Number(d.percent ?? 0);
        mintxt = Number(d.mintxt ?? 1);
        if (Array.isArray(d.linesscore)) {
            let idx = 1;
            linesscore = d.linesscore.map((it) => ({
                idx: idx++,
                score: Number(it?.score ?? 0),
                txt: String(it?.txt ?? ""),
            }));
        }
    }
    else {
        // 回退提取
        const fb = fallbackExtract(raw, notes);
        errcode = fb.errcode;
        errmsg = fb.errmsg;
        percent = fb.percent;
        mintxt = fb.mintxt;
        linesscore = fb.linesscore;
        parser = "relaxed";
    }
    // 判定逻辑
    let decision = "pass";
    const lowScoreSegments = linesscore.filter(x => x.score < minLineScore);
    const next_actions = [];
    const suggestions = [];
    if (mintxt !== 0 || errcode !== "0") {
        decision = "fail";
        next_actions.push(mintxt !== 0 ? "increase_content_length_min_100" : "retry_request_later");
        suggestions.push(mintxt !== 0 ? "正文不足100字，请补充事实与细节后再检测" : "接口繁忙或权限受限，请稍后重试");
    }
    else if (percent > reviewPercent) {
        decision = "fail";
        next_actions.push("remove_template_phrases", "add_verifiable_details", "split_long_sentences");
        suggestions.push("AI占比较高，请去模板化并加入可核验事实，再次检测");
    }
    else if (percent > maxPercent || lowScoreSegments.length > 0) {
        decision = "review";
        if (percent > maxPercent)
            next_actions.push("remove_template_phrases", "add_verifiable_details");
        if (lowScoreSegments.length > 0)
            next_actions.push("revise_intro_reduce_cliches", "split_long_sentences");
        suggestions.push("存在边界风险段落，请按建议最小改写后复验");
    }
    const pass = decision === "pass";
    return {
        errcode,
        errmsg,
        mintxt,
        percent,
        linesscore,
        pass,
        decision,
        next_actions,
        suggestions,
        diagnostics: { parser, raw, notes },
    };
}
//# sourceMappingURL=aiContentDetect.js.map