Twitter/X Research — Topics Mode (v1)

Detects the top trending topics in a category, creates [topic-research].md files, and optionally generates Shorts.

Command

/twitter-research topics:<category> audience:<target audience> [limit:<n>] [shorts:<on|off>]

Examples:

• /twitter-research topics:futbol mexico audience:aficionados liga mx shorts:on limit:10
• /twitter-research topics:champions league audience:fans europeos limit:5

Preconditions

• Chrome is running with --remote-debugging-port=9222
• User is logged into X/Twitter in Chrome
• Backend running for short generation (only if shorts:on):

  curl -s -o /dev/null -w "%{http_code}" http://localhost:8080/api/models
  

Workflow

Step 1 — Discover Trending Topics

Find what is actually trending right now in the target category/region.

1. X trending data (country-specific):

   chrome-devtools navigate_page --url "https://getdaytrends.com/es/mexico/"
   

   Extract the current trend list via evaluate_script. Filter to topics relevant to the category (for futbol: team names, player names, tournament names, transfers, results, goals, referees).
2. X Explore for global/sports trends:

   chrome-devtools navigate_page --url "https://x.com/explore"
   

   Read the Sports / Entertainment / News trend tabs.
3. WebSearch the category + region for today's news:
   • WebSearch("<category> <region> noticias hoy")
   • WebSearch("<category> trending <region> hoy")
   • For futbol: WebSearch("futbol liga mx hoy"), WebSearch("champions league hoy"), WebSearch("mls noticias hoy")
4. Curate the top N topics (default 5, max 10). Choose topics that are:
   • Currently hot (trending in the last 24-48h)
   • Category-relevant
   • Distinct (no duplicates)
   • Video-friendly (visual content available)

Step 2 — Research Each Topic

For each of the top N topics:

1. Get news context via WebSearch: WebSearch("<topic> noticias") and read the top 2-3 results. Capture: what happened, why it matters, key numbers, reactions, official statements.
2. Search X for engagement data:

   chrome-devtools navigate_page --url "https://x.com/search?q=<URL_ENCODED_TOPIC>&src=typed_query&f=top"
   

   Wait 5s, then extract the top 5 posts with real metrics using the scrape extractor (handle, tweet text, status URL, and the numeric bar = replies / retweets / likes / views). Record the single best-performing post URL + its exact stats.
3. Collect media/video links: from the tweet media, news articles, or a quick platform search (WebSearch("site:youtube.com <topic>"), WebSearch("site:tiktok.com <topic>")). Gather 3-5 usable links.

Step 3 — Create [topic-research].md Files

For each topic, create a file named topic-research-<slug>.md in .twitter-research/<YYYY-MM-DD_HHMMSS>/ alongside topics-report.md. The file MUST contain these 6 sections (in this order; section 6 is filled in Step 5):

# Topic Research: <Topic Name>
**Fecha de investigación:** <YYYY-MM-DD>
**Nivel de calor (hotness):** 🔥-🔥🔥🔥🔥🔥

## 1. Título del tema y título viral
**Tema:** <short summary of the topic>
**Títulos virales para video:**
- <viral title 1>
- <viral title 2>

## 2. Contexto de la noticia
<2-4 paragraphs: what happened, why it's hot, key facts, reactions>

## 3. Enlaces de video e imágenes
- <link with description>
- <link with description>

## 4. Guiones de video corto

### Guion A — VIRALIDAD
> [HOOK 0-3s] <hook sentence>
> <4-8 more short TTS-read sentences, one per line>
> [CTA] <call to action>

### Guion B — SHOCK VALUE
> [HOOK] <hook>
> <short punchy lines for TTS>
> <closing line>

## 5. Publicación de referencia + interacción
**Post principal:** <url>
- Respuestas: X | Reposts: X | Me gusta: X | Reproducciones: X
**¿Qué tan caliente está?** <hotness rating + note>

## 6. Título SEO para YouTube
**Título recomendado:** <title> (<score>/10)
**Alternativo A:** <angle> — <title> (<score>/10)
**Alternativo B:** <angle> — <title> (<score>/10)
**Keywords para descripción:** <3-5 search phrases>
**Por qué funciona:** <rules hit from youtube-title>

Script writing rules (for TTS / short vertical video):

• Spanish (or the user's language), spoken-style, no complex punctuation
• Short sentences (6-12 words), one idea per line
• Hook in first 3 seconds (question, bold claim, or shock stat)
• Rhythmic pacing; numbers written out naturally
• End with a CTA (comment / save / follow)
• NO stage directions inside the read text — use [HOOK], [CTA] markers only as labels; the text after them is what the TTS reads

Step 4 — Write topics-report.md

Create an index file listing all topics, their hotness, and file links:

# Topics Report — <category>
**Fecha:** <date>
**Audiencia objetivo:** <audience>

| # | Tema | Calor | Archivo |
|---|------|-------|---------|
| 1 | ... | 🔥🔥🔥 | topic-research-<slug>.md |

Step 5 — Generate SEO YouTube Title (subskill: youtube-title)

For each topic file, load the subskill and produce the best SEO-targeted, attention-grabbing YouTube title:

1. Load it: skill youtube-title
2. Generate 5 candidates from the topic's script hook + key facts, scored on SEO keyword, ALL-CAPS power word, specificity (score/names), stake, and length (45–70 chars) against the channel's own CTR/views data.
3. Keep the best title + 2 A/B alternates and write them into section 6 of the topic file (see template), plus 3-5 description keywords.
4. These titles are the ones used later for the Shorts/YouTube upload.

Why this step exists: titles are the #1 lever on the channel's own data (same video, different title = 13× views). Never skip it.

Step 6 — Video Ideas Pipeline (subskill: video-idea-research)

Turn the research into a publishable slate of ideas: the top 10 videos by views in the window plus 5 video ideas drawn from them, each with 3 scripts and full YouTube metadata.

1. Load it: skill video-idea-research and run with the topic + platform:x (window default month: week|month|year).
2. Rank the topic's video posts by views inside the window (open each status page for the real count — never invent numbers), keep the top 10, and extract each one's outline (hook, beats, CTA, format, duration).
3. Draw 5 ideas from the top-10 outlines (each idea ≥2 source videos).
4. For each idea write 3 TTS-ready scripts — viral (viral_shorts), storytelling (storytelling), inspiration (motivational) — plus the YouTube metadata: title (youtube-title rubric), description, tags, keywords. Backend seeding when available:

   curl -s -X POST http://localhost:8080/api/script \
     -H "Content-Type: application/json" \
     -d '{"videoSubject":"<idea title>","extraPrompt":"<angle + outline context>","aiModel":"g4f","scriptTemplate":"viral_shorts","scriptLength":"standard"}'
   

5. Save video-ideas.json in the session folder, then render the report:

   node "$CLAUDE_SKILL_ROOT/scripts/build-ideas-report.js" .twitter-research/<timestamp>/video-ideas.json
   # fallback: node ./skills/video-idea-research/scripts/build-ideas-report.js <path>
   

   → ideas-report.md + REPORT.html (top 10 videos + 5 ideas).

Step 7 — Optional: Generate Shorts (shorts:on)

For each topic, hand off to the short-generator skill:

1. Load it: skill short-generator
2. Subject = the topic; script = the Virality guion from the topic file
3. Provide hashtags: from the topic keywords and sources: (default instagram,facebook,tiktok,youtube)
4. Collect the video links from section 3 of the topic file and use them in short-generator Step 4/5 (fallback sources if search finds <5 URLs)
5. Show the generated script for user approval before generating the video
6. On approval, generate the Short and download the videos from the links
7. Use the section 6 recommended YouTube title as the video/upload title

Important: Always show the script and get explicit user approval before calling the final video generation endpoint.

Step 8 — Report Completion

✅ Twitter/X Research Topics Mode Complete!

Category: <category>
Audience: <audience>
Topics researched: <N>
  - <topic 1> (🔥hotness) → topic-research-<slug>.md
  - ...
Video ideas: 5 per topic (3 scripts each + YouTube metadata) → REPORT.html
Shorts generated: <M/N>

Next step (optional): skill twitter-research-report for a full HTML report