Threads Growth Skill

Skill by ara.so — Marketing Skills collection.

Data-driven Threads content optimization that learns from your actual post performance. Uses browser automation to scrape Insights, identifies winning patterns, and drafts algorithm-compliant posts with AI-tone detection.

What this does

• Performance tracking: Scrapes your real Threads Insights (views, likes, replies, shares)
• Pattern detection: Identifies what works for YOUR account (not generic advice)
• Smart drafting: Creates posts using your patterns + algorithm red-line checks + AI-tone detection
• Automated updates: Bi-weekly cron keeps your performance database current
• Share optimization: Prioritizes DM forwardability (3-5x weight vs likes per Mosseri 2025)

Installation

# Clone into Hermes skills directory
git clone https://github.com/krumjahn/threads-growth-skill ~/.hermes/skills/social-media/threads-growth

# OR create directory structure manually
mkdir -p ~/.hermes/skills/social-media/threads-growth/{references,scripts}

First-time setup

1. Export Threads cookies

Required for authentication (cookies last ~60 days):

# Install Cookie-Editor browser extension
# Navigate to threads.net (logged in)
# Export cookies as JSON
# Save to ~/.hermes/skills/social-media/threads-growth/cookies.json

2. Run setup script

cd ~/.hermes/skills/social-media/threads-growth
bash scripts/setup.sh

This will:

• Validate cookie format
• Test authentication
• Run first Insights scrape (~5 min)
• Create references/winning-formats.md
• Set up bi-weekly cron job

3. Configure environment

# Set your Threads username
export THREADS_USERNAME="your_username"

# Optional: Custom data path
export THREADS_DATA_PATH="$HOME/.hermes/skills/social-media/threads-growth"

Key files

threads-growth/
├── SKILL.md                        # This file
├── cookies.json                    # Your session cookies (gitignored)
├── references/
│   └── winning-formats.md          # Performance database (auto-updated)
└── scripts/
    ├── setup.sh                    # One-time setup
    └── scrape_insights.sh          # Insights scraper

Core workflows

Draft a new post

# User says: "draft a threads post about [topic]"
# Skill loads winning-formats.md
# Generates draft following your patterns
# Runs algorithm + AI-tone checks
# Returns scored draft

Example pattern matching:

# From winning-formats.md (auto-generated)
Top format: Anti-Demo Hook + Problem/Solution + Proof
Avg engagement: 2,340 views, 186 likes, 23 replies, 12 shares

Structure:
1. Failure state (specific)
2. What you discovered
3. How to implement (3-5 steps)
4. Proof/result (concrete metric)

Best performing: Posts with DM-forwardable utility (how-to, tool lists, templates)

Algorithm red-line check

12 hard suppression rules from Meta's 2025 guidelines:

# Automatically checked in every draft:
# 1. No external links in first post (only replies OK)
# 2. No bare URLs (use "link in bio" instead)
# 3. No follower begging ("follow me", "help me hit X followers")
# 4. No engagement bait ("comment if you agree", "double tap")
# 5. No fake news/misinformation signals
# 6. No sensationalized language (all caps, excessive punctuation)
# 7. No violent/graphic imagery
# 8. No sexual content
# 9. No hate speech
# 10. No spam patterns (repetitive, copy-paste)
# 11. No impersonation
# 12. No coordinated manipulation

AI-tone detection

Flags and fixes 15 common AI tells before posting:

# Sentence-level tells:
- "Delve", "utilize", "leverage" → Replace with plain language
- "It's worth noting that" → Cut
- "In today's digital landscape" → Cut or rephrase
- "Game-changer", "unlock" → Use concrete verbs
- Passive voice → Convert to active

# Structure-level tells:
- Lists starting "1. First..." → Vary openings
- Three parallel paragraphs → Break rhythm
- Summary ending "In conclusion..." → Cut
- Overly balanced pros/cons → Take a stance
- Academic hedging ("perhaps", "somewhat") → Be direct

Score a draft

# Before publishing, get predictive score:
# - DM forwardability (0-10)
# - Repost likelihood (0-10)
# - Algorithm compliance (pass/fail)
# - AI-tone flags (count)
# - Estimated reach multiplier (vs your baseline)

Insights scraper

Manual run

cd ~/.hermes/skills/social-media/threads-growth
bash scripts/scrape_insights.sh

What it collects

# For each post (last 50):
- Post ID
- Publish timestamp
- Text content
- Views (reach)
- Likes
- Replies (total + depth)
- Shares (DM forwards + reposts)
- Save count

Auto-update cron

# Installed during setup:
# Runs every 2 weeks (Sunday 2 AM)
0 2 * * 0 [ $(($(date +\%s) / 604800 \% 2)) -eq 0 ] && bash ~/.hermes/skills/social-media/threads-growth/scripts/scrape_insights.sh

Hook framework

Built-in post structures ranked by performance:

Anti-Demo example structure

I wasted 6 weeks posting at "optimal times"
(my reach actually dropped 40%)

Then I found this:

Threads doesn't care about your timezone.
It cares about YOUR audience's active hours.

How to find them:
1. Export your Insights (this skill does it)
2. Map reply times (not post times)
3. Post 1 hour before peak reply window
4. Track for 2 weeks

My reach jumped 2.3x in 10 days.
Same content. Better timing.

Signal weighting

Based on Mosseri 2025 algorithm updates:

# DM forwards: 3-5x weight of likes
# Deep replies (3+ back-and-forth): 2x weight of top-level
# Saves: 2x weight of likes
# Reposts: 1.5x weight of likes
# Likes: 1x baseline
# Views: Outcome, not signal

Optimization priority:

1. DM forwardability (utility, tools, templates)
2. Conversation depth (ask questions, provide incomplete info)
3. Save-worthiness (reference value)
4. Repost-ability (identity signal for sharer)
5. Likability (lowest priority)

Multi-part Threads

For posts >500 chars:

# Skill automatically chains parts
# Each part: 450-500 chars (leaves buffer)
# Structure:
# Part 1: Hook + problem
# Part 2-N: Steps/details
# Final part: Proof/CTA

# Algorithm treats as one post if:
# - Published within 60 seconds
# - No external engagement between parts
# - Same thread chain

Configuration options

# Environment variables (optional)
export THREADS_USERNAME="your_username"          # Required
export THREADS_DATA_PATH="$HOME/.hermes/..."    # Default shown
export THREADS_SCRAPE_LIMIT=50                   # Posts to fetch (default 50)
export THREADS_MIN_PERFORMANCE_SAMPLE=10         # Min posts for pattern detection
export THREADS_COOKIE_WARNING_DAYS=7             # Days before expiry to warn

Troubleshooting

Cookie expiry

# Symptoms: Scraper fails with 401/403
# Fix: Re-export cookies from browser
# 1. Login to threads.net
# 2. Export cookies (Cookie-Editor)
# 3. Save to cookies.json
# 4. Run: bash scripts/scrape_insights.sh

Missing Insights data

# Threads only shows Insights for posts with 100+ views
# For new accounts: Wait until you have 10+ qualifying posts
# Check: references/winning-formats.md for sample size

# If sample size < 10:
# - Patterns unreliable
# - Skill uses general best practices until enough data

Scraper hanging

# Usually: Slow network or Threads rate limiting
# Fix 1: Kill process, wait 5 min, retry
# Fix 2: Reduce THREADS_SCRAPE_LIMIT to 25

pkill -f scrape_insights
sleep 300
bash scripts/scrape_insights.sh

Pattern detection not working

# Need minimum viable sample:
# - 10+ posts with Insights
# - At least 5 with 500+ views
# - Published over 2+ weeks

# Check your data:
cat references/winning-formats.md | grep "Sample size"

Algorithm knowledge base

Adapted from AK-Threads-Booster:

Boost signals

• Original thinking (not reposted content)
• Conversation starters (questions, incomplete info)
• DM-forwardable utility
• Identity-signal repostability
• Consistent posting (3-5x/week better than daily)

Suppression signals

• External links in main post
• Engagement bait
• Follower begging
• Repetitive content
• Low reply rate (<2%)

Timing myths

• "Post at 9 AM" → Wrong. Post 1hr before YOUR audience's active window
• "Post daily" → Wrong. 3-5x/week with high quality beats daily low quality
• "First hour decides reach" → Partially true. First 3 hours matter most.

Example: Complete workflow

# 1. User requests
"Draft a threads post about my new AI coding workflow"

# 2. Skill loads patterns
cat references/winning-formats.md
# Identifies: Anti-Demo hook + 3-step process performs best for you

# 3. Generates draft
I tried using ChatGPT for code review.
It missed 3 critical bugs in production code.

Switched to Claude + this prompt structure:
1. Show the code
2. Ask "what breaks first?"
3. Request refactor with error handling

Caught 12 issues ChatGPT missed.
Same AI. Better questions.

# 4. Runs checks
✓ Algorithm compliance (no red-line violations)
✓ AI-tone scan (0 flags)
✓ DM forwardability: 8/10 (actionable steps)
✓ Estimated reach: 1.4x your baseline

# 5. User approves and posts

Data privacy

• All data stored locally (~/.hermes/skills/social-media/threads-growth)
• Cookies never transmitted (used only for Threads authentication)
• Insights data never shared
• No tracking, no analytics

Performance baseline

Typical results after 6 weeks:

• 1.5-2.5x reach improvement
• 2-3x engagement rate increase
• 40-60% reduction in low-performing posts
• Better pattern recognition every 2-week cycle

Updates

# Pull latest version
cd ~/.hermes/skills/social-media/threads-growth
git pull origin main

# Refresh patterns (optional)
bash scripts/scrape_insights.sh

---

Built by @krumjahn — Running live since April 2026. MIT License.