feat: content engine

packages/content-engine/src/generator.ts

@@ -1,5 +1,5 @@
 import OpenAI from "openai";
-import { ResearchAgent, Fact, SocialPost } from "@mintel/journaling";
+import { ResearchAgent, type Fact, type SocialPost } from "@mintel/journaling";
 import { MemeGenerator, MemeSuggestion } from "@mintel/meme-generator";
 import * as fs from "node:fs/promises";
 import * as path from "node:path";
@@ -237,11 +237,21 @@ REGELN:
         console.log(`   → ${factInsertions.length} fact enrichments planned`);
       }
 
-      // ----- STEP 1.5: Social Media Search -----
-      console.log("📱 Identifying real social media posts...");
-      const socialPosts = await this.researchAgent.findSocialPosts(
-        content.substring(0, 200),
-      );
+      // ----- STEP 1.5: Social Media Extraction (no LLM — regex only) -----
+      console.log("📱 Extracting existing social media embeds...");
+      const socialPosts = this.researchAgent.extractSocialPosts(content);
+
+      // If none exist, fetch real ones via Serper API
+      if (socialPosts.length === 0) {
+        console.log(
+          "   → None found. Fetching real social posts via Serper API...",
+        );
+        const newPosts = await this.researchAgent.fetchRealSocialPosts(
+          content.slice(0, 500),
+        );
+        socialPosts.push(...newPosts);
+      }
+
       if (socialPosts.length > 0) {
         console.log(
           `📝 Planning placement for ${socialPosts.length} social media posts...`,
@@ -593,7 +603,7 @@ RULES:
   - youtube -> <YouTubeEmbed videoId="ID" />
   - twitter -> <TwitterEmbed tweetId="ID" theme="light" />
   - linkedin -> <LinkedInEmbed urn="ID" />
-- Add a 1-sentence intro paragraph above the embed to contextualize it.
+- Add a 1-sentence intro paragraph above the embed to contextualize it naturally in the flow of the text (e.g. "Wie Experte XY im folgenden Video detailliert erklärt:"). This context is MANDATORY. Do not just drop the Component without text reference.
 
 CONTEXT:
 ${context.slice(0, 3000)}
@@ -842,6 +852,11 @@ Tone: ${tone}.
 Facts: ${factsContext}
 ${componentsContext}
 
+BLOG POST BEST PRACTICES (MANDATORY):
+- DEVIL'S ADVOCATE: Füge zwingend eine kurze kritische Sektion ein (z.B. mit \`<ComparisonRow>\` oder \`<IconList>\`), in der du offen die Nachteile/Kosten/Haken deiner eigenen Lösung ansprichst ("Der Haken an der Sache...").
+- FAQ GENERATOR: Am absoluten Ende des Artikels erstellst du zwingend eine Markdown-Liste mit den 3 wichtigsten Fragen (FAQ) und Antworten (jeweils 2 Sätze) für Google Rich Snippets.
+- Nutze wo passend die obigen React-Komponenten für ein hochwertiges Layout.
+
 Format as Markdown. Start with # H1.
 For places where a diagram would help, insert: <!-- DIAGRAM_PLACEHOLDER: Concept Name -->
 Return ONLY raw content.`,
@@ -891,6 +906,7 @@ RULES:
 - CRITICAL: Generate ONLY ONE single connected graph. Do NOT generate multiple independent graphs or isolated subgraphs in the same Mermaid block.
 - No nested subgraphs. Keep instructions short.
 - Use double-quoted labels for nodes: A["Label"]
+- VERY CRITICAL: DO NOT use curly braces '{}' or brackets '[]' inside labels unless they are wrapped in double quotes (e.g. A["Text {with braces}"]). 
 - VERY CRITICAL: DO NOT use any HTML tags (no <br>, no <br/>, no <b>, etc).
 - VERY CRITICAL: DO NOT use special characters like '&', '<', '>', or double-quotes inside the label strings. They break the mermaid parser in our environment.
 - Return ONLY the raw mermaid code. No markdown blocks, no backticks.