ct-alpha

# CT Alpha — Crypto Twitter Intelligence

Turn X/Twitter into an actionable crypto intelligence layer. Search CT for narratives, alpha, strategies, and sentiment, then rank results using TweetRank (a PageRank-inspired credibility scoring system), extract tokens/CAs from multiple signals, detect coordinated raids, and suggest execution steps using available tools.

## Overview

- **Search**: Query CT with crypto-optimized noise filters and relevancy sorting
- **TweetRank**: Score tweets by author credibility + engagement quality + recency
- **Multi-Signal Token Detection**: Cashtags, name-phrases, crypto URLs, contract addresses
- **Raid Detection**: Flag tickers promoted mostly by low-credibility accounts
- **Trending**: Detect trending tokens across multiple search queries
- **Watchlist**: Monitor trusted CT accounts by category
- **Thread Hydration**: Fetch full conversation threads
- **X Articles**: Full-text extraction of long-form posts (>280 chars)
- **Cost Tracking**: Per-session API spend tracking (~$0.005/tweet)
- **Dynamic Tool Discovery**: Suggest follow-up actions with DeFi Llama, Backpack, Polymarket, etc.

## Prerequisites

- **Runtime**: [Bun](https://bun.sh) (TypeScript runtime)
- **API Token**: X API Bearer Token from [developer.x.com](https://developer.x.com) (pay-per-use via xAI)
- **Cost**: ~$0.005 per tweet read, ~$0.10 per quick search (20 tweets)

## Quick Start

### Installation

```bash
# Clone the skill
git clone https://github.com/yashhsm/skills.git
cd skills/skills/ct-alpha

# Run the installer (configures token, watchlist, cache)
bun run install.ts
```

### Environment Setup

```bash
# Set your X API Bearer Token
export X_BEARER_TOKEN="your_token_here"

# Or save to persistent env file
mkdir -p ~/.config/env
echo 'export X_BEARER_TOKEN="your_token_here"' >> ~/.config/env/global.env
source ~/.config/env/global.env
```

### Basic Usage

```bash
# Search for alpha on a token
bun run ct-search.ts search "$SOL alpha" --quick

# Detect trending tokens
bun run ct-search.ts trending --window 6h --solana-only

# Monitor watchlist accounts
bun run ct-search.ts watchlist --since 24h

# Read a specific tweet or article
bun run ct-search.ts read https://x.com/user/status/123456

# Check API spending
bun run ct-search.ts cost
```

## CLI Reference

### search — Core research command

```bash
bun run ct-search.ts search "<query>" [flags]
```

| Flag | Description | Default |
|------|-------------|---------|
| `--quick` | 20 tweets, 1hr cache, ~$0.10 | **Default mode** |
| `--full` | Up to 100 tweets, 15min cache, ~$0.50 | Confirm cost first |
| `--limit N` | Max total tweets | 20 (quick), 100 (full) |
| `--sort <field>` | `likes`, `recency`, `relevancy` | `relevancy` |
| `--since <duration>` | `1h`, `6h`, `24h`, `7d` | `24h` |
| `--min-likes N` | Engagement filter | 3 (quick) |
| `--from user1,user2` | Restrict to specific accounts | — |
| `--extract-tickers` | Show extracted tickers | — |
| `--extract-cas` | Show contract addresses and crypto URLs | — |
| `--raw` | JSON output | — |

### trending — Multi-signal trending detection

```bash
bun run ct-search.ts trending [flags]
```

| Flag | Description | Default |
|------|-------------|---------|
| `--window <duration>` | `1h`, `6h`, `24h` | `6h` |
| `--min-mentions N` | Minimum mention count | 3 |
| `--solana-only` | Solana ecosystem only | — |
| `--top N` | Top N results | 20 |

### watchlist — Monitor CT accounts

```bash
bun run ct-search.ts watchlist [flags]
```

| Flag | Description | Default |
|------|-------------|---------|
| `--category <cat>` | Filter by category | all |
| `--since <duration>` | Time window | `24h` |

### read — Read a specific tweet/article

```bash
bun run ct-search.ts read <tweet_url_or_id> [--thread] [--raw]
```

Accepts x.com URLs, twitter.com URLs, or raw tweet IDs. Articles (long-form posts) are fetched in full.

### thread — Hydrate conversation thread

```bash
bun run ct-search.ts thread <tweet_id>
```

### cost — Track API spending

```bash
bun run ct-search.ts cost [--reset]
```

## Core Features

### TweetRank Scoring

Every tweet is scored by three factors multiplied together:

1. **AuthorCred (0-10)**: Watchlist membership (+5), follower/following ratio (capped +5), verification (+1), account age (log scale, capped +2), bot penalty (-3)
2. **EngagementQuality**: Bookmarks (×3, unfakeable) > Quotes (×2.5, high effort) > Likes (×1.5) > Retweets (×1, easily botted). All log-scaled.
3. **RecencyBoost**: `1 / (1 + hoursAgo / 24)` — newer tweets score higher.

```
TweetRank = AuthorCred × EngagementQuality × RecencyBoost
```

Each tweet receives a source label:
- `[WATCHLIST]` — Author is on your watchlist (highest trust)
- `[HIGH-CRED]` — AuthorCred ≥ 5
- `[UNKNOWN]` — Unverified author
- `[SUSPICIOUS]` — Bot-like patterns detected

### Multi-Signal Token Detection

Cashtag-only detection misses how tokens actually spread on CT. This extracts from four signal types:

| Signal | Example | Confidence |
|--------|---------|------------|
| **Cashtag** | `$SOL`, `$JTO` | High |
| **Name-phrase** | "pendle" + crypto context | High/Medium |
| **Crypto URL** | pump.fun, dexscreener, birdeye, jup.ag | High |
| **Contract Address** | Base58 (Solana) / 0x (Ethereum) with context | High/Low |

Supported crypto URL domains: pump.fun, dexscreener.com, birdeye.so, jup.ag, raydium.io, solscan.io, etherscan.io.

### Raid Detection

Detects coordinated pump campaigns by analyzing author credibility distribution per ticker:

- If >70% of authors mentioning a ticker have low credibility (AuthorCred < 3), it's flagged as a potential raid
- Output includes raid score, total/low-cred author counts

### Noise Filtering

Every search auto-appends crypto noise filters:
- `-is:retweet` (removes retweets)
- `-"airdrop"`, `-"giveaway"`, `-"whitelist"` (spam removal)
- `-"follow and RT"`, `-"follow & RT"`, `-"free mint"`, `-"dm to claim"` (engagement bait)
- Quick mode also adds `-is:reply`

### Caching

Aggressive