Facebook Research — Main Skill (v2.3)

Full workflow: Setup → Scrape → Analyze → Report → Engage → Template

Quick Start

skill facebook-research-setup       # Step 1: Setup (prerequisites + config + competitors)
skill facebook-research-scrape      # Step 2: Scrape page + competitors analytics
skill facebook-research-analyze     # Step 3: Deep deconstruct top 5 + competitor videos
skill facebook-research-report      # Step 4: REPORT.html with history trends + competitor analysis
skill facebook-research-engage      # Step 5: Generate engagement opportunities (10+ at a time)
skill facebook-research-grow        # Step 5b: Grow followers — goal-driven "audience stealer" campaigns (e.g. +200 in 24h)
skill facebook-research-template    # Step 6: Download reels + transcribe + analyze + generate template

Architecture

skills/facebook-research/          # Skill (bundled, global-installable via ~/.claude/skills/)
├── SKILL.md
└── scripts/                       # Bundled — no external .fb-research/scripts needed
    ├── package.json               # Node deps (chrome-remote-interface)
    ├── scrape.js                  # Page + competitor scraper via Chrome DevTools
    ├── report-html.js             # HTML report generator
    ├── download-reels.js          # Top-5 reel downloader via Chrome DevTools
    └── transcribe.sh              # Whisper transcription of downloaded reels

.fb-research/                      # Tool data directory (project root, auto-resolved via CWD walk-up)
└── projects/<page-name>/
    ├── config.json                       # Project config (page + competitors with URLs)
    ├── page-analytics-history.json       # All historical snapshots (global, appended each run)
    ├── report-history.json               # Historical report snapshots (global)
    ├── replication-history.json          # History of all replication scripts (global)
    ├── latest.json                       # Pointer to latest sessionId
    ├── competitors/                      # Legacy mirror of last run (for back-compat)
    │   ├── <competitor-1>/raw-posts.json
    │   └── <competitor-2>/raw-posts.json
    └── <YYYY-MM-DD_HHMMSS>/             # One folder per run (dated)
        ├── page-analytics.json           # This run's page stats
        ├── raw-posts.json                # This run's raw posts
        ├── competitors/<name>/raw-posts.json
        ├── discovered-competitors.json
        ├── top5-analysis.md
        ├── REPORT.html                   # Self-contained report for this run
        ├── report.md
        ├── replication-scripts/
        ├── screenshots/                  # Hook screenshots
        ├── screenshots-competitors/
        ├── screenshots-engagement/       # Screenshots of target posts + top comments
        ├── screenshots-growth/           # Screenshots of target posts + our visitor comments
        ├── growth-campaign.json
        ├── growth-targets.json
        ├── growth-log.json
        ├── growth-progress.json
        ├── growth-strategy.md
        ├── video-template/
        │   ├── downloads/                # Downloaded reels for this run
        │   ├── transcripts/
        │   ├── transcript-analysis.json
        │   ├── VIDEO-TEMPLATE.md
        │   └── best-performer-breakdown.md
        └── engagement-opportunities.json

Execution model (hybrid): scraping, reporting, reel download, and transcription run as bundled scripts in skills/facebook-research/scripts/ (Chrome DevTools Protocol). Skills resolve the data dir via CWD walk-up (./.fb-research), FB_RESEARCH_ROOT env, or --data-root. Dated runs live in projects/<name>/<YYYY-MM-DD_HHMMSS>/. Use the Chrome DevTools CLI directly for quick DOM reads, comment posting, and screenshots. Global-install safe (Muse ~/.claude/skills/ — Opencode also reads it).

Video Template Pipeline

This pipeline deconstructs the page's top 5 performing reels down to their DNA and builds a reusable template.

How It Works

1. Download: Chrome DevTools navigates to each of the top 5 post insight pages, extracts the video source URL from the <video> element, and downloads the MP4 file.
2. Transcribe: Whisper (local) processes each downloaded video and generates transcriptions (.txt) with timestamps.
3. Analyze: All 5 transcripts are analyzed for:
   • Hook pattern: First 5 seconds — what's the exact phrasing, tone, and pacing?
   • Content structure: How is the narrative organized? (Tease → Context → Drama → CTA)
   • Pacing: Sentence length, pause points, question frequency, emphasis words
   • Vocabulary: Key phrases, power words, repetition patterns
   • CTA structure: How is the call-to-action framed? Question format?
   • Emotional arc: How does the video progress emotionally scene by scene
   • Length & timing: Exact timing of each section
4. Template: The #1 performing video (by views/engagement) is used as the reference. A reusable fill-in-the-blank template is generated.

Video Template Format (VIDEO-TEMPLATE.md)

The template is a structured, fillable document with:

## Template: [PATTERN NAME]
**Source**: [original post title] — [views] views

### Hook (0:00-0:03)
Text overlay: "[FORMULA]"
Audio: "[EXACT WORDING PATTERN]"

### Context (0:03-0:10)
Structure: [pattern of how context is established]
Vocabulary: [power words used]
Pacing: [sentence rhythm]

### Drama/Body (0:10-0:40)
Narrative flow: [story structure]
Emotional arc: [how emotions shift]
Key phrases: [repeatable phrases]

### CTA (0:40-0:50)
Format: [question formula]
Hook type: [debate / opinion / prediction]
Length: [seconds]

### Production Specs
- Duration: [seconds]
- Text overlay style: [font/color/position]
- Audio style: [music type + commentary]
- Transition speed: [fast/medium/slow]
- Caption style: [styled / auto]

Mapping New Content to the Template

Your Topic: [fill in]
Hook:    ¡[STAR] [VERB] [TEAM]! [EMOJI]
Context: [match situation] + [stakes]
Drama:   [key moment 1] → [reaction] → [key moment 2] → [climax]
CTA:     [controversial question] + ¡Te leo en los comentarios!

Commands

Engagement Opportunity Engine (v3)

The engagement system visits competitor pages, scrapes their top posts and comments, and generates snarky/rage-bait comments that match the audience's real voice.

How It Works

1. Visits each competitor page via Chrome DevTools
2. Finds their most recent popular posts (sorted by engagement)
3. Navigates into each post to get the real permalink URL
4. Extracts the top 10 most-liked comments from the post
5. Analyzes comment patterns: tone, slang (Mexican for Spanish pages), inside jokes
6. Generates 3 comment variants per post:
   • Hot take: Strong controversial opinion
   • Sarcastic zinger: Short, meme-like jab
   • Analysis + jab: Shows knowledge then takes a shot
7. Takes screenshots of the post and comments section
8. Saves everything with real post URLs for one-click action

Key Rules

• Match the audience's tone — if they're sarcastic, be sarcastic. Don't sound like a bot.
• Use Mexican slang for Spanish pages: "no mames", "wey", "qué pedo", "chingón", "neta", "carnal"
• Don't promote yourself — comments should look like they're from a passionate fan
• Rage-bait triggers: bad reffing, overrated players, controversial takes, hypocrisy
• 1-3 sentences max — short comments get more likes than long rants

Engagement Hook Formula

[SPECIFIC REFERENCE to post] + [OPINIONATED TAKE] + [MEXICAN SLANG / INSIDER JOKE] + [JAB or RHETORICAL QUESTION]

Comment Templates (Spanish with Mexican Slang)

JSON Schema (engagement-opportunities.json)

{
  "id": 1,
  "batch": 1,
  "createdAt": "YYYY-MM-DD HH:MM:SS",
  "competitorName": "...",
  "competitorUrl": "https://www.facebook.com/...",
  "postUrl": "https://www.facebook.com/...",
  "postPreview": "...",
  "screenshot": "screenshots-engagement/...png",
  "commentsScreenshot": "screenshots-engagement/...comments.png",
  "topComments": [
    {"commenter": "...", "text": "...", "likes": "5K"}
  ],
  "commentPatterns": {
    "tone": "sarcastic",
    "slangUsed": ["wey", "no mames"],
    "commonThemes": ["bad reffing", "player x is overrated"]
  },
  "generatedComments": [
    {"variant": 1, "type": "hot_take", "text": "...", "whyItGetsLikes": "..."},
    {"variant": 2, "type": "sarcastic_zinger", "text": "...", "whyItGetsLikes": "..."},
    {"variant": 3, "type": "analysis_jab", "text": "...", "whyItGetsLikes": "..."}
  ],
  "value": "high",
  "used": false
}

Notes

• Requires Chrome logged into Facebook (Cookies imported via /setup-browser-cookies or manual login)
• Video download requires being logged into Facebook (videos are gated behind authentication)
• Whisper is required: pip install openai-whisper
• ffmpeg is required for audio extraction if video download fails
• All output paths are relative to dataRoot (resolved from CWD's .fb-research, or FB_RESEARCH_ROOT)
• Each run creates projects/<name>/<YYYY-MM-DD_HHMMSS>/ and appends a snapshot to page-analytics-history.json (global) for trend tracking
• User can ask for engagement opportunities ANY time, not just in the engage state