Reddit Research — Generate State (v3)

Takes scraped Reddit posts (with viral scripts already embedded), finds CONTEXT-RELEVANT videos, and generates Shorts.

CSV for MagicSync: after generating Shorts (ask the user), build the bulk-scheduling CSV with node "$CLAUDE_SKILL_ROOT/scripts/build-csv.js" <timestamp> (see the main reddit-research skill, Phase 3).

Key Improvement over v1

Video search queries must be specific to the post topic, not generic. Example:

• Post about Klopp → search "Jürgen Klopp German national team new era coach 2026"
• Post about Mbappé Paraguay → search "Mbappé Paraguay penalty controversy dark arts 2026"
• Post about referee → search "Ilgiz Tantashev referee Paraguay France no cards 2026"

Before Running

1. Ensure .reddit-research/<timestamp>/ exists with scraped posts (from Phase 1)
2. Ensure Flask backend is running on port 8080
3. Chrome DevTools available

Steps

1. Select Research Session

ls -d .reddit-research/*/

Read the index.json to get the post list and select the most recent session.

2. For Each Post, Search for Context-Relevant Videos

For each post, construct a targeted search query based on the post title and content:

Use Chrome DevTools on DuckDuckGo Videos

chrome-devtools navigate_page --url "https://duckduckgo.com/"
chrome-devtools fill "<search_input_uid>" "<specific_search_query>"
chrome-devtools press_key "Enter"

Wait for results. Click the "Videos" tab:

chrome-devtools take_snapshot
# Find the Videos tab/link and click it
chrome-devtools click "<videos_tab_uid>"

Then extract video URLs (YouTube, TikTok, Instagram, etc.):

() => {
  const links = document.querySelectorAll('a[href*="youtube.com/watch"], a[href*="youtu.be"], a[href*="tiktok.com"], a[href*="instagram.com/reel"], a[href*="vimeo.com"], a[href*="dailymotion.com"]');
  const results = [];
  const seen = new Set();
  links.forEach(a => {
    const href = a.getAttribute('href');
    let actualUrl = href;
    try {
      const u = new URL(href);
      if (u.hostname.includes('duckduckgo') && u.searchParams.get('uddg')) {
        actualUrl = decodeURIComponent(u.searchParams.get('uddg'));
      }
    } catch(e) {}
    if (actualUrl && !seen.has(actualUrl)) {
      seen.add(actualUrl);
      results.push({ title: a.innerText?.trim() || a.getAttribute('title') || '', url: actualUrl });
    }
  });
  return JSON.stringify(results.slice(0, 8));
}

Fallback: WebSearch

If Chrome DevTools yields fewer than 3 video URLs:

websearch "<specific_search_query> video"

Extract video URLs from results. Prefer YouTube, TikTok, Instagram Reel URLs.

3. Append Sourced Videos to Post Markdown File

Read the post's markdown file. Find the ## Sourced Videos section (or add it). Append the discovered video URLs:

## Sourced Videos

1. [Video Title](<video_url>)
2. [Video Title](<video_url>)
...

4. Call short-generator for Each Post

The markdown file already contains:

• Title
• Post content
• Top comments
• Viral video script (in Spanish)
• Sourced video URLs

To generate the Short, call the short-generator:

/short-generator <post_title> voice:M5 hashtags:<keyword1,keyword2> sources:youtube,tiktok,instagram ratio:9:16 music:off subtitle_position:center,bottom

When the short-generator asks for:

• Script: Use the script already saved in the ## Viral Video Script (Spanish) section
• Video URLs: Provide the URLs from ## Sourced Videos
• Voice: M5 (Daniel)
• Subtitle position: center,bottom
• Language: Spanish
• Hashtags: Extract keywords from the post title (e.g., Klopp, Germany, football)

5. Track Generation Status

Append to each post's markdown file:

## Short Generation

- **Status**: ✅ Generated / ❌ Failed
- **Voice**: M5
- **Language**: Spanish
- **Subtitle Position**: center,bottom
- **Generated At**: <YYYY-MM-DD HH:MM:SS>

7. Generate from the Video Ideas Pipeline (video-ideas.json)

When the session has video-ideas.json (from the video-idea-research playbook), each idea is already production-ready:

1. Read .reddit-research/<timestamp>/video-ideas.json.
2. Ask the user which idea(s) to produce (rank order by view estimate).
3. Use the idea's scripts.viral as the script (or storytelling / inspiration per the user's pick) — paste-ready, pure spoken text.
4. Use the idea's mediaLinks as the sourced video URLs (Step 2 search is then optional — links already collected).
5. Upload title/description/tags/keywords come from the idea's youtube metadata block — never invent new ones.
6. Then run Step 4's short-generator call with those values.

8. Report Completion

✅ Shorts Generation Complete!

Session: .reddit-research/<timestamp>/
Posts processed: 10/10
Generated: X
Failed: Y

Video files:
- http://localhost:8080/static/generated_videos/<file1>.mp4
- http://localhost:8080/static/generated_videos/<file2>.mp4
...