Testing Image Analysis

This guide explains how to verify that the image analysis feature is working correctly when monitoring Twitter timelines.

Prerequisites

1. OpenAI API Key: Ensure OPENAI_API_KEY is set in your .env file

2. Twitter Authentication: Run bun run auth:twitter if not already authenticated

3. Database Access: Supabase credentials configured

Testing Methods

Method 1: Check Current Status

Run the check script to see current image analysis status:

bun run agent/plugins/twitter/scripts/check-image-analysis.ts

This shows:

• Total tweets with images

• How many were AI-analyzed

• Screenshot detection stats

• Account-specific analysis

Method 2: Manual Timeline Monitoring

Trigger timeline monitoring for a specific account:

# Monitor Elon Musk's timeline (often has images)
bun run agent/plugins/twitter/scripts/trigger-timeline-monitoring.ts elonmusk

# Monitor Trump's timeline (Truth Social screenshots)
bun run agent/plugins/twitter/scripts/trigger-timeline-monitoring.ts realDonaldTrump

# Monitor any account
bun run agent/plugins/twitter/scripts/trigger-timeline-monitoring.ts [username]

This will:

1. Fetch recent tweets from the account

2. Process any images found

3. Show you the analysis results

Method 3: Database Query

Run SQL queries directly in Supabase (see agent/plugins/twitter/sql/check-image-analysis.sql for more queries):

-- Check recent image analysis
SELECT 
  source_account,
  SUBSTRING(text, 1, 50) as tweet_text,
  CASE 
    WHEN image_description = 'Image attached to tweet' THEN '❌ Not analyzed'
    WHEN image_description LIKE '%[Extracted text:%' THEN '✅ Screenshot'
    ELSE '✅ Analyzed'
  END as status,
  SUBSTRING(image_description, 1, 100) as description
FROM timeline_tweets
WHERE image_url IS NOT NULL
ORDER BY cached_at DESC
LIMIT 20;

Method 4: Watch Timeline Monitoring Logs

When the Twitter plugin is running, it monitors timelines every 30 minutes. Watch the logs:

# In your main application logs, look for:
🤖 Analyzing X images with AI...
✅ Image analysis complete: Screenshot detected

Understanding Results

Successful AI Analysis

When image analysis is working, you'll see descriptions like:

"The image shows a graph depicting Bitcoin price movements over the past month 
[Topics: cryptocurrency, bitcoin, trading]"

Or for screenshots:

"Screenshot of a Truth Social post by Donald Trump 
[Extracted text: 'The Fake News Media...'] 
[Source: Truth Social] 
[Topics: politics, media]"

Default Description

If you see "Image attached to tweet", it means:

• Image wasn't analyzed (no OpenAI key or error)

• Analysis failed

• Image already had alt text

Troubleshooting

1. All images show "Image attached to tweet"

   • Check: Is OPENAI_API_KEY set?

   • Run: echo $OPENAI_API_KEY to verify

   • Test: bun run agent/plugins/twitter/tests/ImageAnalysisService.test.ts

2. Some images fail

   • Check logs for "Error analyzing image"

   • Common issues:

     • Image URL not accessible

     • Image too large

     • Rate limiting

3. No tweets with images found

   • Monitor accounts that post images frequently

   • Try: elonmusk, NASA, NatGeo

Expected Performance

• Analysis Time: 2-3 seconds per image

• Batch Processing: 2 images concurrently

• Rate Limiting: 1.5s delay between batches

• Cache: Previously analyzed images are cached

Testing Image Analysis Service Directly

# Run unit tests
bun run agent/plugins/twitter/tests/ImageAnalysisService.test.ts

# Run integration test
bun run agent/plugins/twitter/tests/timeline-image-integration.test.ts

Monitoring in Production

1. Regular Checks: Run check script daily

2. SQL Monitoring: Use provided SQL queries

3. Log Analysis: Monitor for analysis errors

4. Cost Tracking: Monitor OpenAI API usage

Important Accounts to Monitor

For best testing results, monitor these accounts:

• @realDonaldTrump: Truth Social screenshots

• @elonmusk: Memes and charts

• @naval: Quote images

• @NatGeo: High-quality photos

• @NASA: Space images

Success Metrics

Image analysis is working well when:

• ✅ 80%+ of images have AI descriptions

• ✅ Screenshots are detected with text extraction

• ✅ Topics and sentiment are identified

• ✅ No repeated API errors in logs