autocrew 0.1.0

package/src/tools/research.ts

@@ -0,0 +1,304 @@
+import path from "node:path";
+import fs from "node:fs/promises";
+import { Type } from "@sinclair/typebox";
+import { saveTopic } from "../storage/local-store.js";
+import { browserCdpAdapter } from "../adapters/browser/browser-cdp.js";
+import { researchWithTikHub } from "../adapters/research/tikhub.js";
+import { runFreeResearch, type SearchResult } from "../modules/research/free-engine.js";
+import type { BrowserPlatform, ResearchItem } from "../adapters/browser/types.js";
+
+type ResearchMode = "auto" | "browser_first" | "api_fallback" | "free" | "manual";
+
+export const researchSchema = Type.Object({
+  action: Type.Unsafe<"discover" | "session_status">({
+    type: "string",
+    enum: ["discover", "session_status"],
+    description: "Research action: discover new topics or inspect browser session status.",
+  }),
+  industry: Type.Optional(Type.String({ description: "Industry or niche. Falls back to MEMORY.md if omitted." })),
+  keyword: Type.Optional(Type.String({ description: "Research keyword or angle." })),
+  platform: Type.Optional(
+    Type.Unsafe<BrowserPlatform>({
+      type: "string",
+      enum: ["xiaohongshu", "douyin", "wechat_mp", "wechat_video", "bilibili"],
+      description: "Target platform.",
+    }),
+  ),
+  topic_count: Type.Optional(Type.Number({ description: "How many topics to create. Default: 3." })),
+  save_topics: Type.Optional(Type.Boolean({ description: "Save generated topics into ~/.autocrew/topics. Default: true." })),
+  mode: Type.Optional(
+    Type.Unsafe<ResearchMode>({
+      type: "string",
+      enum: ["auto", "browser_first", "api_fallback", "free", "manual"],
+      description: "Execution mode. 'free' uses web search + viral scoring (no browser/API needed). Default: auto.",
+    }),
+  ),
+  search_results: Type.Optional(
+    Type.Array(
+      Type.Object({
+        title: Type.String(),
+        snippet: Type.String(),
+        url: Type.String(),
+      }),
+      { description: "Pre-fetched search results for 'free' mode. Caller provides these from web_search." },
+    ),
+  ),
+});
+
+interface MemoryContext {
+  industry?: string;
+  competitors: string[];
+}
+
+function resolveDataDir(customDir?: string): string {
+  if (customDir) return customDir;
+  const home = process.env.HOME || process.env.USERPROFILE || "~";
+  return path.join(home, ".autocrew");
+}
+
+function normalizePlatform(platform?: string): BrowserPlatform {
+  return (platform as BrowserPlatform) || "xiaohongshu";
+}
+
+async function readMemoryContext(dataDir?: string): Promise<MemoryContext> {
+  const target = path.join(resolveDataDir(dataDir), "MEMORY.md");
+  try {
+    const raw = await fs.readFile(target, "utf-8");
+    const competitors = raw
+      .split("\n")
+      .map((line) => line.trim())
+      .filter((line) => line.startsWith("- "))
+      .map((line) => line.replace(/^- /, ""))
+      .filter((line) => /xhs|小红书|douyin|抖音|bilibili|b站|wechat/i.test(line));
+
+    const industryMatch =
+      raw.match(/industry[:：]\s*(.+)/i) ||
+      raw.match(/定位[:：]\s*(.+)/) ||
+      raw.match(/行业[:：]\s*(.+)/);
+
+    return {
+      industry: industryMatch?.[1]?.trim(),
+      competitors,
+    };
+  } catch {
+    return { competitors: [] };
+  }
+}
+
+function truncateTitle(text: string, max = 20): string {
+  const compact = text.replace(/\s+/g, "").trim();
+  if (compact.length <= max) return compact;
+  return `${compact.slice(0, max - 1)}…`;
+}
+
+function buildTopicTitle(item: ResearchItem, platform: BrowserPlatform): string {
+  const base = item.title || `${platform}选题`;
+  return truncateTitle(base);
+}
+
+function buildTopicDescription(
+  item: ResearchItem,
+  industry: string,
+  platform: BrowserPlatform,
+): string {
+  const summary = item.summary?.trim() || "来自浏览器登录态下的近期内容观察";
+  const sourceLabel =
+    item.source === "browser_cdp" ? "浏览器登录态" : item.source === "api_provider" ? "API fallback" : "人工降级";
+  return `${summary}。适合 ${industry || "当前账号"} 在 ${platform} 上继续延展，来源：${sourceLabel}${item.author ? `，参考账号：${item.author}` : ""}。`;
+}
+
+function buildTopicTags(industry: string, platform: BrowserPlatform): string[] {
+  const tags = [platform];
+  if (industry) {
+    tags.push(industry);
+  }
+  if (platform === "xiaohongshu") {
+    tags.push("选题");
+  }
+  if (platform === "douyin") {
+    tags.push("短视频");
+  }
+  return Array.from(new Set(tags));
+}
+
+async function runDiscovery(params: Record<string, unknown>) {
+  const dataDir = (params._dataDir as string) || undefined;
+  const memory = await readMemoryContext(dataDir);
+  const industry = ((params.industry as string) || memory.industry || "").trim();
+  const platform = normalizePlatform(params.platform as string | undefined);
+  const keyword = ((params.keyword as string) || industry || "内容选题").trim();
+  const topicCount = Number(params.topic_count || 3);
+  const saveTopics = params.save_topics !== false;
+  const mode = (params.mode as ResearchMode) || "auto";
+
+  let items: ResearchItem[] = [];
+  const sourcesUsed: string[] = [];
+
+  // --- Free mode: web search + viral scoring engine ---
+  if (mode === "free") {
+    const searchResults = (params.search_results as SearchResult[] | undefined) || [];
+    const freeResult = await runFreeResearch({
+      keyword,
+      searchResults,
+      topicCount,
+      dataDir,
+    });
+
+    const topics = freeResult.candidates.map((c) => ({
+      title: c.title,
+      description: c.description,
+      tags: c.tags,
+      source: c.source,
+    }));
+
+    const saved = [];
+    if (saveTopics) {
+      for (const topic of topics) {
+        saved.push(await saveTopic(topic, dataDir));
+      }
+    }
+
+    return {
+      ok: true,
+      mode: "free",
+      platform,
+      keyword,
+      industry: freeResult.industry,
+      competitors: memory.competitors,
+      sourcesUsed: ["free_engine"],
+      topics: saveTopics ? saved : topics,
+      savedCount: saveTopics ? saved.length : 0,
+      searchQueries: freeResult.searchQueries,
+      filtersApplied: freeResult.filtersApplied,
+      summary: freeResult.summary,
+      candidates: freeResult.candidates,
+    };
+  }
+
+  // --- Browser-first mode (Pro path) ---
+  if (mode === "auto" || mode === "browser_first") {
+    items = await browserCdpAdapter.research({ platform, keyword, limit: topicCount });
+    if (items.length > 0) {
+      sourcesUsed.push(browserCdpAdapter.id);
+    }
+  }
+
+  if (items.length === 0 && (mode === "auto" || mode === "api_fallback")) {
+    items = await researchWithTikHub({ platform, keyword, limit: topicCount });
+    if (items.length > 0) {
+      sourcesUsed.push("tikhub_fallback");
+    }
+  }
+
+  // --- Auto fallback to free engine when browser + API both empty ---
+  if (items.length === 0 && mode === "auto") {
+    const searchResults = (params.search_results as SearchResult[] | undefined) || [];
+    if (searchResults.length > 0) {
+      const freeResult = await runFreeResearch({
+        keyword,
+        searchResults,
+        topicCount,
+        dataDir,
+      });
+      if (freeResult.candidates.length > 0) {
+        const topics = freeResult.candidates.map((c) => ({
+          title: c.title,
+          description: c.description,
+          tags: c.tags,
+          source: c.source,
+        }));
+
+        const saved = [];
+        if (saveTopics) {
+          for (const topic of topics) {
+            saved.push(await saveTopic(topic, dataDir));
+          }
+        }
+
+        return {
+          ok: true,
+          mode: "free",
+          platform,
+          keyword,
+          industry: freeResult.industry,
+          competitors: memory.competitors,
+          sourcesUsed: ["free_engine_auto_fallback"],
+          topics: saveTopics ? saved : topics,
+          savedCount: saveTopics ? saved.length : 0,
+          searchQueries: freeResult.searchQueries,
+          filtersApplied: freeResult.filtersApplied,
+          summary: freeResult.summary,
+          candidates: freeResult.candidates,
+        };
+      }
+    }
+  }
+
+  if (items.length === 0) {
+    items = Array.from({ length: topicCount }).map((_, index) => ({
+      title: `${keyword} 选题方向 ${index + 1}`,
+      summary: "手动降级模式生成，请后续结合真实平台反馈再筛一次。",
+      platform,
+      source: "manual",
+    }));
+    sourcesUsed.push("manual");
+  }
+
+  const topics = items.slice(0, topicCount).map((item) => ({
+    title: buildTopicTitle(item, platform),
+    description: buildTopicDescription(item, industry, platform),
+    tags: buildTopicTags(industry, platform),
+    source: `${item.source}:${platform}`,
+  }));
+
+  const saved = [];
+  if (saveTopics) {
+    for (const topic of topics) {
+      saved.push(await saveTopic(topic, dataDir));
+    }
+  }
+
+  return {
+    ok: true,
+    mode: mode === "auto" ? "browser_first" : mode,
+    platform,
+    keyword,
+    industry: industry || null,
+    competitors: memory.competitors,
+    sourcesUsed,
+    topics: saveTopics ? saved : topics,
+    savedCount: saveTopics ? saved.length : 0,
+    note:
+      "Current browser-first adapter is a structural runtime entry. Replace its placeholder implementation with host-specific CDP/web-access execution next.",
+  };
+}
+
+async function getSessionStatuses(params: Record<string, unknown>) {
+  const requestedPlatform = params.platform as BrowserPlatform | undefined;
+  const platforms: BrowserPlatform[] = requestedPlatform
+    ? [requestedPlatform]
+    : ["xiaohongshu", "douyin", "wechat_mp", "wechat_video", "bilibili"];
+
+  const items = [];
+  for (const platform of platforms) {
+    if (browserCdpAdapter.getSessionStatus) {
+      items.push(await browserCdpAdapter.getSessionStatus(platform));
+    }
+  }
+
+  return {
+    ok: true,
+    sessions: items,
+  };
+}
+
+export async function executeResearch(params: Record<string, unknown>) {
+  const action = (params.action as string) || "discover";
+  if (action === "discover") {
+    return runDiscovery(params);
+  }
+  if (action === "session_status") {
+    return getSessionStatuses(params);
+  }
+  return { ok: false, error: `Unknown action: ${action}` };
+}

package/src/tools/review.ts

@@ -0,0 +1,305 @@
+/**
+ * autocrew_review tool — content review integrating:
+ * 1. Sensitive word scanning
+ * 2. De-AI check (humanizer-zh dry-run)
+ * 3. Quality scoring (info density, hook strength, CTA clarity, readability)
+ * 4. Auto-fix (apply suggestions)
+ *
+ * PRD §6: "autocrew review 命令整合敏感词检测"
+ */
+import { Type } from "@sinclair/typebox";
+import { scanText, type ScanResult } from "../modules/filter/sensitive-words.js";
+import { humanizeZh, type HumanizeZhResult } from "../modules/humanizer/zh.js";
+import { getContent, updateContent, transitionStatus, normalizeLegacyStatus } from "../storage/local-store.js";
+
+// --- Types ---
+
+export interface QualityScore {
+  /** 0-100 overall */
+  total: number;
+  /** Sub-scores, each 0-25 */
+  infoDensity: number;
+  hookStrength: number;
+  ctaClarity: number;
+  readability: number;
+  /** Per-dimension notes */
+  notes: string[];
+}
+
+export interface ReviewReport {
+  ok: boolean;
+  /** Did the content pass all checks? */
+  passed: boolean;
+  sensitiveWords: ScanResult;
+  aiCheck: { hasAiTraces: boolean; changeCount: number; changes: string[] };
+  qualityScore: QualityScore;
+  /** Combined summary */
+  summary: string;
+  /** Suggested fixes */
+  fixes: string[];
+  /** Auto-fixed text (if available) */
+  autoFixedText?: string;
+}
+
+// --- Schema ---
+
+export const reviewSchema = Type.Object({
+  action: Type.Unsafe<"full_review" | "scan_only" | "quality_score" | "auto_fix">({
+    type: "string",
+    enum: ["full_review", "scan_only", "quality_score", "auto_fix"],
+    description:
+      "Action: 'full_review' runs all checks, 'scan_only' runs sensitive word scan only, " +
+      "'quality_score' returns quality score only, 'auto_fix' applies all auto-fixable suggestions and saves.",
+  }),
+  content_id: Type.Optional(Type.String({ description: "AutoCrew content id to review." })),
+  text: Type.Optional(Type.String({ description: "Raw text to review directly (if no content_id)." })),
+  platform: Type.Optional(Type.String({ description: "Target platform for platform-specific checks." })),
+});
+
+// --- Quality Scoring ---
+
+function scoreQuality(text: string, platform?: string): QualityScore {
+  const notes: string[] = [];
+
+  // --- Info Density (0-25) ---
+  // Penalize filler, reward concrete data points
+  const charCount = text.length;
+  const sentences = text.split(/[。！？\n]/).filter((s) => s.trim().length > 0);
+  const avgSentenceLen = charCount / Math.max(sentences.length, 1);
+  const dataPoints = (text.match(/\d+[%％万亿个条次天月年]/g) || []).length;
+  let infoDensity = 15; // baseline
+  if (dataPoints >= 3) infoDensity += 5;
+  else if (dataPoints >= 1) infoDensity += 2;
+  if (avgSentenceLen > 80) {
+    infoDensity -= 5;
+    notes.push("句子偏长，建议拆分");
+  }
+  if (avgSentenceLen < 15 && sentences.length > 3) {
+    infoDensity += 3;
+  }
+  // Penalize filler phrases
+  const fillerCount = (text.match(/值得一提|需要注意|综上所述|总而言之|可以说/g) || []).length;
+  if (fillerCount > 0) {
+    infoDensity -= Math.min(5, fillerCount * 2);
+    notes.push(`发现 ${fillerCount} 处套话，建议删除`);
+  }
+  infoDensity = clamp(infoDensity, 0, 25);
+
+  // --- Hook Strength (0-25) ---
+  const firstLine = sentences[0]?.trim() || "";
+  let hookStrength = 10;
+  // Question hook
+  if (/[？?]/.test(firstLine)) {
+    hookStrength += 5;
+    notes.push("开头用了提问式 hook ✓");
+  }
+  // Number hook
+  if (/\d/.test(firstLine)) {
+    hookStrength += 4;
+  }
+  // Emotional trigger
+  if (/别再|千万|后悔|真相|没想到|居然|竟然|震惊|绝了/.test(firstLine)) {
+    hookStrength += 5;
+  }
+  // Short punchy opening
+  if (firstLine.length <= 20 && firstLine.length > 0) {
+    hookStrength += 3;
+  }
+  // Penalty: generic opening
+  if (/大家好|今天我们|在当今社会|随着.*的发展/.test(firstLine)) {
+    hookStrength -= 8;
+    notes.push("开头太泛，建议用具体场景或数据切入");
+  }
+  hookStrength = clamp(hookStrength, 0, 25);
+
+  // --- CTA Clarity (0-25) ---
+  const lastThird = text.slice(Math.floor(text.length * 0.7));
+  let ctaClarity = 10;
+  // Explicit CTA
+  if (/关注|收藏|点赞|转发|评论|私信|留言|试试|赶紧|快去/.test(lastThird)) {
+    ctaClarity += 8;
+    notes.push("结尾有明确 CTA ✓");
+  }
+  // Question CTA
+  if (/[？?]/.test(lastThird)) {
+    ctaClarity += 4;
+  }
+  // No CTA at all
+  if (ctaClarity === 10) {
+    notes.push("结尾缺少 CTA，建议加引导互动的句子");
+  }
+  // Platform-specific CTA
+  if (platform === "xhs" || platform === "xiaohongshu") {
+    if (/收藏|关注/.test(lastThird)) ctaClarity += 3;
+  }
+  if (platform === "douyin") {
+    if (/关注|点赞/.test(lastThird)) ctaClarity += 3;
+  }
+  ctaClarity = clamp(ctaClarity, 0, 25);
+
+  // --- Readability (0-25) ---
+  let readability = 15;
+  // Paragraph count
+  const paragraphs = text.split(/\n{2,}/).filter((p) => p.trim());
+  if (paragraphs.length >= 3 && paragraphs.length <= 8) {
+    readability += 3;
+  }
+  // Emoji usage
+  const emojiCount = (text.match(/[\u{1F300}-\u{1F9FF}]/gu) || []).length;
+  if (emojiCount >= 1 && emojiCount <= 10) {
+    readability += 3;
+  } else if (emojiCount > 15) {
+    readability -= 3;
+    notes.push("emoji 过多，建议精简");
+  }
+  // Line breaks
+  if (paragraphs.some((p) => p.length > 300)) {
+    readability -= 5;
+    notes.push("有段落超过 300 字，建议拆分");
+  }
+  // Short paragraphs bonus
+  const shortParas = paragraphs.filter((p) => p.length <= 100).length;
+  if (shortParas / Math.max(paragraphs.length, 1) > 0.5) {
+    readability += 4;
+  }
+  readability = clamp(readability, 0, 25);
+
+  const total = infoDensity + hookStrength + ctaClarity + readability;
+
+  return { total, infoDensity, hookStrength, ctaClarity, readability, notes };
+}
+
+function clamp(v: number, min: number, max: number): number {
+  return Math.max(min, Math.min(max, v));
+}
+
+// --- Execute ---
+
+export async function executeReview(params: Record<string, unknown>) {
+  const action = (params.action as string) || "full_review";
+  const dataDir = (params._dataDir as string) || undefined;
+  const platform = (params.platform as string) || undefined;
+  const contentId = params.content_id as string | undefined;
+
+  // Resolve text
+  let text = (params.text as string) || "";
+  let title = "";
+  if (!text && contentId) {
+    const content = await getContent(contentId, dataDir);
+    if (!content) return { ok: false, error: `Content ${contentId} not found` };
+    text = content.body;
+    title = content.title;
+  }
+  if (!text) return { ok: false, error: "text or content_id is required" };
+
+  const fullText = title ? `${title}\n\n${text}` : text;
+
+  // --- scan_only ---
+  if (action === "scan_only") {
+    const scanResult = await scanText(fullText, platform, dataDir);
+    return { ok: true, action, sensitiveWords: scanResult };
+  }
+
+  // --- quality_score ---
+  if (action === "quality_score") {
+    const score = scoreQuality(fullText, platform);
+    return { ok: true, action, qualityScore: score };
+  }
+
+  // --- auto_fix ---
+  if (action === "auto_fix") {
+    // 1. Sensitive word auto-fix
+    const scanResult = await scanText(fullText, platform, dataDir);
+    let fixedText = scanResult.autoFixedText || fullText;
+
+    // 2. Humanizer pass
+    const humanResult = humanizeZh({ text: fixedText });
+    fixedText = humanResult.humanizedText;
+
+    // Save back if content_id provided
+    if (contentId) {
+      await updateContent(contentId, { body: fixedText }, dataDir);
+    }
+
+    return {
+      ok: true,
+      action,
+      autoFixedText: fixedText,
+      sensitiveWordsFixed: scanResult.hits.filter((h) => h.suggestion).length,
+      aiFixesApplied: humanResult.changeCount,
+      saved: !!contentId,
+    };
+  }
+
+  // --- full_review ---
+  // 1. Sensitive words
+  const sensitiveWords = await scanText(fullText, platform, dataDir);
+
+  // 2. AI check (dry-run humanizer)
+  const humanResult = humanizeZh({ text: fullText });
+  const aiCheck = {
+    hasAiTraces: humanResult.changeCount > 0,
+    changeCount: humanResult.changeCount,
+    changes: humanResult.changes,
+  };
+
+  // 3. Quality score
+  const qualityScore = scoreQuality(fullText, platform);
+
+  // 4. Build fixes list
+  const fixes: string[] = [];
+  if (sensitiveWords.hitCount > 0) {
+    fixes.push(`修复 ${sensitiveWords.hitCount} 个敏感词`);
+    for (const hit of sensitiveWords.hits.slice(0, 5)) {
+      const fix = hit.suggestion ? `"${hit.word}" → "${hit.suggestion}"` : `删除"${hit.word}"`;
+      fixes.push(`  - ${fix} (${hit.category})`);
+    }
+  }
+  if (aiCheck.hasAiTraces) {
+    fixes.push(`去 AI 味：${aiCheck.changeCount} 处需要修改`);
+  }
+  for (const note of qualityScore.notes) {
+    fixes.push(note);
+  }
+
+  // 5. Determine pass/fail
+  const passed =
+    sensitiveWords.hitCount === 0 &&
+    !aiCheck.hasAiTraces &&
+    qualityScore.total >= 60;
+
+  // 6. Build summary
+  const parts: string[] = [];
+  parts.push(passed ? "✅ 审核通过" : "⚠️ 审核未通过");
+  parts.push(`敏感词: ${sensitiveWords.hitCount === 0 ? "✓ 无" : `✗ ${sensitiveWords.hitCount} 个`}`);
+  parts.push(`AI 痕迹: ${aiCheck.hasAiTraces ? `✗ ${aiCheck.changeCount} 处` : "✓ 无"}`);
+  parts.push(`质量评分: ${qualityScore.total}/100 (信息密度${qualityScore.infoDensity} Hook${qualityScore.hookStrength} CTA${qualityScore.ctaClarity} 可读性${qualityScore.readability})`);
+  const summary = parts.join("\n");
+
+  // 7. Auto-transition if content_id provided
+  if (contentId) {
+    const targetStatus = passed ? "approved" : "revision";
+    const diffNote = passed ? undefined : fixes.join("; ");
+    await transitionStatus(
+      contentId,
+      normalizeLegacyStatus(targetStatus),
+      { diffNote },
+      dataDir,
+    ).catch(() => {
+      /* transition may fail if not in reviewing state — that's ok */
+    });
+  }
+
+  const report: ReviewReport = {
+    ok: true,
+    passed,
+    sensitiveWords,
+    aiCheck,
+    qualityScore,
+    summary,
+    fixes,
+    autoFixedText: sensitiveWords.autoFixedText,
+  };
+
+  return report;
+}