X Algorithm Tweet Optimizer

A CLI tool that transforms any tweet into an algorithm-optimized viral tweet. Uses X's open-sourced "For You" feed algorithm (Phoenix model) to analyze and maximize the 19 engagement signals that determine tweet ranking.

How It Works

X's recommendation algorithm predicts 19 engagement action probabilities for each tweet and ranks them by weighted sum:

Final Score = Σ (weight_i × P(action_i))

This tool runs a two-phase interactive flow:

1. Analyzes your tweet's structural features (hooks, questions, hashtags, power words, etc.)
2. Scores it across all 19 algorithm dimensions using heuristic estimation
3. Optimizes the tweet while preserving your original voice and style
4. Lets you refine — chat with the AI to request specific changes
5. Optionally generates different style variations targeting different signal combinations
6. Copies the final tweet to your clipboard with one keypress

The 19 Signals (and What Matters Most)

Signal	Weight	What Drives It
share_via_dm_score	100.0	Content people DM to friends
quote_score	40.0	Hot takes that invite commentary
reply_score	27.0	Questions and debate triggers
repost_score	10.0	Quotable insights, universal truths
follow_author_score	10.0	Demonstrated expertise
report_score	-9209.0	Policy violations (avoid!)
block_author_score	-371.0	Offensive content (avoid!)

A single DM share is worth 100 Likes. A single report wipes out everything.

Requirements

• Python 3.10+
• Claude Code CLI installed and authenticated
• pyyaml

Setup

pip install pyyaml

Usage

# Basic optimization (interactive mode)
python optimize.py "AI will replace 80% of jobs in 5 years"

# Turkish tweet
python optimize.py "Yapay zeka işlerin %80'ini alacak" --lang tr

# With topic context
python optimize.py "AI will replace 80% of jobs" --topic "AI"

# Casual style for Phase 2 variations
python optimize.py "AI will replace 80% of jobs" --style casual

# From file, thread format
python optimize.py --file draft.txt --thread

# Tweet with media
python optimize.py "Check out this chart" --media --topic "data science"

# Non-interactive mode (legacy, shows all variations at once)
python optimize.py "AI will replace 80% of jobs" --no-interactive --variations 5

# JSON output for programmatic use
python optimize.py "Test tweet" --json

Interactive Flow

By default, the tool runs interactively:

Phase 1 — Same-Style Optimization

Your tweet is optimized while keeping your original voice, structure, and meaning:

  ╔══════════════════════════════════════════════════════════════╗
  ║  X ALGORITHM TWEET OPTIMIZER                                ║
  ╚══════════════════════════════════════════════════════════════╝

  ── ORIGINAL TWEET ────────────────────────────────────────────────

    AI will replace 80% of jobs in 5 years

  Characters: 38/280  │  Lang: EN  │  Score: 20%

  ── OPTIMIZED TWEET  20% → 35% (+15pts) ──────────────────────────

    AI will replace 80% of jobs in 5 years.

    But not the ones you think.

    The question isn't IF — it's which tasks disappear first.

  Characters: 112/280

  What would you like to do?
    [1] Copy optimized tweet to clipboard
    [2] Refine with AI (request changes)
    [3] Generate different style variations
    [4] Copy original tweet to clipboard
    [5] Exit

Refine with AI

Select [2] to chat with the AI and request changes. You can iterate as many times as you want:

  What would you like to change?
  (Type your instructions, e.g. 'make it shorter', 'keep the URL', 'add humor')

  > add a question at the end to trigger replies

  Refining tweet...

  ── OPTIMIZED TWEET  20% → 42% (+22pts) ──────────────────────────

    AI will replace 80% of jobs in 5 years.

    But not the ones you think.

    The question isn't IF — it's which tasks disappear first.

    What job do you think goes first?

  Characters: 147/280

After each refinement, you return to the menu and can refine again, copy, or generate variations.

Phase 2 — Style Variations

Select [3] to generate multiple variations with different strategies (Reply Magnet, DM Share Magnet, Quote Trigger, etc.), then pick one to copy.

CLI Options

Flag	Description	Default
tweet	Tweet text (positional)	-
--topic	Topic/niche context	None
--lang	Language: en, tr, auto	auto
--variations	Number of style variations (Phase 2)	3
--style	Tone: professional, casual, provocative, educational	professional
--media	Tweet will include photo/video	false
--thread	Optimize for thread format	false
--file	Read tweet from file	None
--json	Output as JSON (non-interactive)	false
--verbose	Show all 19 signals in detail	false
--no-interactive	Skip interactive prompts, show all variations directly	false

Configuration

Edit config.yaml to customize:

• Category weights — adjust how Engagement, Discoverability, Shareability, Content Quality, and Safety contribute to the overall score
• Display settings — bar width, number of signals shown, show all vs top-N
• Claude timeout — increase for complex prompts

Project Structure

x_content/
├── optimize.py          # CLI entry point (interactive flow)
├── config.yaml          # Settings
├── pyproject.toml       # Dependencies
└── x_content/
    ├── algorithm.py     # 19 actions, weights, signal map
    ├── analyzer.py      # Structural tweet feature extraction
    ├── scorer.py        # Heuristic scoring + comparison metrics
    ├── prompts.py       # Claude prompt templates (preserve-style, refine, variations)
    ├── optimizer.py     # Claude Code CLI subprocess integration
    ├── display.py       # Terminal output formatting (ANSI colors, dynamic width)
    └── config.py        # YAML config loader

Algorithm Source

Based on X's open-sourced recommendation algorithm:

• phoenix/runners.py — 19 action definitions and ranking logic
• Scoring uses jax.nn.sigmoid(logits) to produce per-action probabilities
• Primary ranking by favorite_score, final ranking by weighted combination