X Research

X/Twitter Research

Research high-performing tweets from tracked accounts, identify outliers, and optionally analyze video content for hooks and structure.

Prerequisites

• APIFY_TOKEN environment variable or in .env
• GEMINI_API_KEY environment variable or in .env (for video analysis)
• apify-client and google-genai Python packages
• Accounts configured in .claude/context/x-accounts.md

Verify setup:

python3 -c "
import os
try:
    from dotenv import load_dotenv
    load_dotenv()
except ImportError:
    pass
from apify_client import ApifyClient
assert os.environ.get('APIFY_TOKEN'), 'APIFY_TOKEN not set'
" && echo "Prerequisites OK"

Workflow

1. Create Run Folder

RUN_FOLDER="x-research/$(date +%Y-%m-%d_%H%M%S)" && mkdir -p "$RUN_FOLDER" && echo "$RUN_FOLDER"

2. Fetch Tweets

python3 .claude/skills/x-research/scripts/fetch_tweets.py \
  --days 30 \
  --max-items 100 \
  --output {RUN_FOLDER}/raw.json

Parameters:

• --days: Days back to search (default: 30)
• --max-items: Max tweets per account (default: 100)
• --handles: Override accounts file with specific handles

API Limits: Minimum 50 tweets per query required. Wait a couple minutes between runs.

⚠️ 注意:

• 使用 searchTerms + from:handle 语法，不要用 twitterHandles（后者不稳定）
• 已自动排除转推 (-filter:retweets)

3. Identify Outliers

python3 .claude/skills/x-research/scripts/analyze_posts.py \
  --input {RUN_FOLDER}/raw.json \
  --output {RUN_FOLDER}/outliers.json \
  --threshold 2.0

Output JSON contains:

• total_posts: Number of tweets analyzed
• outlier_count: Number of outliers found
• topics: Top hashtags, mentions, and keywords
• content_patterns: Analysis of what formats perform well
• accounts: List of accounts analyzed
• outliers: Array of outlier tweets with engagement metrics

4. Analyze Videos with AI (Optional)

If outliers contain video content:

python3 .claude/skills/video-content-analyzer/scripts/analyze_videos.py \
  --input {RUN_FOLDER}/outliers.json \
  --output {RUN_FOLDER}/video-analysis.json \
  --platform x \
  --max-videos 5

Note: X/Twitter is primarily text-based. Video analysis is optional and only useful when outliers contain video posts.

5. Generate Report

Read {RUN_FOLDER}/outliers.json (and optionally {RUN_FOLDER}/video-analysis.json), then generate {RUN_FOLDER}/report.md.

Report Structure:

# X/Twitter Research Report

Generated: {date}

## Summary

- **Total tweets analyzed**: {total_posts}
- **Outlier tweets identified**: {outlier_count}
- **Outlier rate**: {percentage}%

## Top Performing Tweets (Outliers)

### 1. @{username} ({name})

> {tweet_text}

- **URL**: {url}
- **Date**: {created_at}
- **Engagement**: {likes} likes | {retweets} RTs | {replies} replies | {bookmarks} bookmarks
- **Engagement Score**: {score}
- **Engagement Rate**: {rate}%
- **Followers**: {followers}

[Repeat for top 15 outliers]

## Top Performing Hooks (if video analysis available)

### Hook 1: {technique} - @{username}
- **Opening**: "{opening_line}"
- **Why it works**: {attention_grab}
- **Replicable Formula**: {replicable_formula}
- [Watch Video]({url})

## Trending Topics

### Top Hashtags
[From outliers.json topics.hashtags]

### Top Keywords
[From outliers.json topics.keywords]

### Top Mentions
[From outliers.json topics.mentions]

## Content Patterns in Outliers

| Pattern | Count | Percentage |
|---------|-------|------------|
| Contains media | {count} | {pct}% |
| Contains external link | {count} | {pct}% |
| Thread format | {count} | {pct}% |
| Quote tweet | {count} | {pct}% |
| Asks a question | {count} | {pct}% |
| List/numbered format | {count} | {pct}% |
| Short (<100 chars) | {count} | {pct}% |
| Medium (100-200 chars) | {count} | {pct}% |
| Long (>200 chars) | {count} | {pct}% |

## Actionable Takeaways

[Synthesize patterns into 4-6 specific recommendations]

## Accounts Analyzed

[List accounts]

Focus on actionable insights. Content patterns and trending topics are key for X/Twitter research.

Quick Reference

Full pipeline:

RUN_FOLDER="x-research/$(date +%Y-%m-%d_%H%M%S)" && mkdir -p "$RUN_FOLDER" && \
python3 .claude/skills/x-research/scripts/fetch_tweets.py -o "$RUN_FOLDER/raw.json" && \
python3 .claude/skills/x-research/scripts/analyze_posts.py -i "$RUN_FOLDER/raw.json" -o "$RUN_FOLDER/outliers.json"

With video analysis (optional):

python3 .claude/skills/video-content-analyzer/scripts/analyze_videos.py -i "$RUN_FOLDER/outliers.json" -o "$RUN_FOLDER/video-analysis.json" -p x

Then read JSON files and generate the report.

Engagement Metrics

Engagement Score (weighted):

• Bookmarks: 4x (highest signal - saved for reference)
• Replies: 3x (active conversation)
• Retweets: 2x (amplification)
• Quotes: 2x (engagement with commentary)
• Likes: 1x (passive approval)

Outlier Detection: Tweets with engagement rate > mean + (threshold x std_dev)

Engagement Rate: (score / followers) x 100

Output Location

All output goes to timestamped run folders:

x-research/
└── {YYYY-MM-DD_HHMMSS}/
    ├── raw.json            # Raw tweet data from Apify
    ├── outliers.json       # Outliers with metadata and topics
    ├── video-analysis.json # AI video analysis (optional)
    └── report.md           # Final report