Agent Skillsleamsigc/ShortsGenerator › facebook-research-template

facebook-research-template

GitHub

自动下载Facebook热门Reels,通过Whisper转录并分析结构模式,生成可复用视频模板及报告,助力内容创作者复制成功策略。

skills/facebook-research/facebook-research-template/SKILL.md leamsigc/ShortsGenerator

Trigger Scenarios

需要分析社交媒体热门视频结构 基于数据生成标准化视频创作模板

Install

npx skills add leamsigc/ShortsGenerator --skill facebook-research-template -g -y
More Options

Non-standard path

npx skills add https://github.com/leamsigc/ShortsGenerator/tree/main/skills/facebook-research/facebook-research-template -g -y

Use without installing

npx skills use leamsigc/ShortsGenerator@facebook-research-template

指定 Agent (Claude Code)

npx skills add leamsigc/ShortsGenerator --skill facebook-research-template -a claude-code -g -y

安装 repo 全部 skill

npx skills add leamsigc/ShortsGenerator --all -g -y

预览 repo 内 skill

npx skills add leamsigc/ShortsGenerator --list

SKILL.md

Frontmatter
{
    "name": "facebook-research-template",
    "parent": "facebook-research",
    "version": "1.0.0",
    "description": "Download the top 5 best-performing reels from a scraped Facebook page,\ntranscribe them with Whisper, analyze the transcripts for structural\npatterns, and generate a reusable video template based on the best\nperformer so the user can replicate or exceed its results."
}

facebook-research-template — Video Template Extraction

Full pipeline: Download → Transcribe → Analyze → Template → Update REPORT.html

Prerequisites

  • openai-whisper installed (pip install openai-whisper)
  • ffmpeg installed
  • Chrome logged into Facebook (needed to access video pages)
  • Previous states completed (scrape + analyze must have run)

Pipeline Steps

Step 1: Download Top 5 Reels

node .fb-research/scripts/download-reels.js <project-name>

The script navigates to each of the top 5 posts (from page-analytics.json top5), extracts the <video> src, and downloads to video-template/downloads/post-N.mp4. If src is a blob URL or download fails, fall back to Chrome DevTools manual capture of the network request that serves the video.

Step 2: Transcribe with Whisper

bash .fb-research/scripts/transcribe.sh <project-name>

Processes every file in video-template/downloads/ with Whisper base model (Spanish) and writes .txt + .srt to video-template/transcripts/. Skips already-transcribed files. Adjust --model in the script for speed (tiny) or accuracy (large).

Step 3: Analyze Transcripts

For each of the 5 transcripts, extract:

  1. Hook (0-5s): Exact wording, timing, tone
  2. Context (5-15s): How the situation is established
  3. Body/Drama (15-40s): The narrative arc, key moments
  4. CTA (40-50s): Call to action format
  5. Pacing: Sentence length distribution, questions per video, emphasis patterns
  6. Vocabulary: Power words, repeated phrases, emotional triggers
  7. Emotional arc: How the video makes the viewer feel moment-by-moment

Compare all 5 to find:

  • What the best performer (#1) does differently
  • Common patterns across all top performers
  • What the worst performer (#5) is missing

Step 4: Generate Reusable Template

Create a fill-in-the-blank template based on the #1 post's structure.

The template includes:

  • Exact hook formula with slots: ¡[STAR NAME] + [DRAMATIC VERB] + [TEAM]! [EMOJI]
  • Scene-by-scene structure with timing
  • Vocabulary list (power words that work)
  • CTA formula
  • Production specs (duration, text style, audio)

Step 5: Update REPORT.html

Add a new section "🎬 Video Template" to REPORT.html showing:

  • The reusable template
  • The best performer breakdown
  • Comparison table of all 5 transcripts

Output Files

File Description
video-template/downloads/post-1.mp4 through post-5.mp4 Downloaded reels
video-template/transcripts/post-1.txt through post-5.txt Whisper transcriptions
video-template/transcripts/post-1.srt through post-5.srt Whisper subtitles with timestamps
video-template/transcript-analysis.json Structured analysis of all 5 transcripts
video-template/VIDEO-TEMPLATE.md Reusable fill-in-the-blank video template
video-template/best-performer-breakdown.md Deep analysis of the winner

Video Template Format

# Video Template: [Pattern Name]

**Based on**: [Post Title] — [views] views · [engagement] engagement
**Pattern**: [star+drama / upset-alert / curiosity-gap]

## 1. Hook Formula (0:00-0:03)

¡[STAR_NAME] [DRAMATIC_VERB] [TEAM/COUNTRY]! [EMOJI]

**Power words that work**:
- SALVA, ELIMINA, ROMPE, DESTROZA, HACE MAGIA
- INCREÍBLE, IMPOSIBLE, NADIE LO ESPERABA
- POLÉMICA, ROBO, INJUSTICIA, SOPRESA

## 2. Scene Structure

| Time | Section | Content | Emotion |
|------|---------|---------|---------|
| 0:00-0:03 | Hook | Bold text overlay + dramatic audio | Shock/Curiosity |
| 0:03-0:10 | Context | "Team was [situation]. But then..." | Tension |
| 0:10-0:30 | Drama | Key moment + reaction + analysis | Peak emotion |
| 0:30-0:45 | Opinion | "The real story is..." → controversial take | Debate |
| 0:45-0:50 | CTA | Question + subscribe prompt | Engagement |

## 3. Narrative Arc

1. **Pattern interrupt** (0s): Bold claim that stops scroll
2. **Context hook** (3s): "Everyone thought X, but Y happened"
3. **Escalation** (10s): Building the drama step by step
4. **Climax** (25s): The key moment described vividly
5. **Resolution** (35s): Your hot take/analysis
6. **Engagement bait** (45s): Question that splits the audience

## 4. Caption Formula

¡[HOOK]! [EMOJI]

[2-3 sentences: context + drama + opinion]

[CONTROVERSIAL QUESTION]
¿Crees que...? ¡Te leo en los comentarios! 👇

[hashtags: 2 viral + 2 niche + 1 branded]

## 5. Production Specs

- Duration: 45-50 seconds
- Text overlay: Bold upper-case, white with black stroke
- Audio: Match commentary + cinematic background music
- Captions: Styled (bold_outline or modern_glow)
- Transitions: Fast cuts (1-2 second clips)
- Color: High contrast, warm tones for excitement

## 6. Fill-in-the-Blanks Template

Copy this for each new video:

Hook: ¡[] [] [____]! [emoji] Context: [Team] estaba [situation]. Pero entonces... Drama: [Key moment] significó [impact] para [team]. Opinion: Lo que nadie dice es [controversial take]. CTA: ¿Crees que [debate question]? ¡Te leo! 👇

Notes

  • If download fails (Facebook blocks), generate the template from the available metadata + analysis data
  • Whisper model can be adjusted: tiny (fast) to large (accurate)
  • For Spanish content, base model works well; use large for accented speech

Version History

  • 7b83166 Current 2026-08-29 02:30

Same Skill Collection

skills/facebook-research-analyze/SKILL.md
skills/facebook-research-engage/SKILL.md
skills/facebook-research-report/SKILL.md
skills/facebook-research-scrape/SKILL.md
skills/facebook-research-setup/SKILL.md
skills/ig-research-analyze/SKILL.md
skills/ig-research-report/SKILL.md
skills/ig-research-scrape/SKILL.md
skills/ig-research-setup/SKILL.md
skills/ig-research-transcribe/SKILL.md
skills/ig-research/SKILL.md
skills/list-workflows/SKILL.md
skills/new-workflow/SKILL.md
skills/onboard/SKILL.md
skills/reddit-research-generate/SKILL.md
skills/reddit-research-scrape/SKILL.md
skills/reddit-research/SKILL.md
skills/short-generator/SKILL.md
skills/twitter-research-analyze/SKILL.md
skills/twitter-research-report/SKILL.md
skills/twitter-research-scrape/SKILL.md
skills/twitter-research-setup/SKILL.md
skills/twitter-research-topics/SKILL.md
skills/workflow-conventions/SKILL.md
skills/facebook-research-grow/SKILL.md
skills/facebook-research/SKILL.md
skills/twitter-research/SKILL.md

Metadata

Files
0
Version
7b83166
Hash
72cc7b0b
Indexed
2026-08-29 02:30

Accueil - Wiki
Copyright © 2011-2026 iteam. Current version is 2.155.2. UTC+08:00, 2026-08-30 11:52
浙ICP备14020137号-1 $Carte des visiteurs$