Priority

🔴 URGENT - HIGHEST

Summary

Multiple bots affected by cache system recursion bug triggered by image processing. Verified cases: prime-heron ($15.09 in 90 minutes) and sharp-otter ($8.58). System generates 26-33 million cache read tokens per image.

Verified Impact

Prime-heron (Jinbot):

Image processing period: 2026-02-26 05:40-07:10 (90 minutes)
Historical consumption: $15.30 (normal usage over days)  
Image bug consumption: $15.09 (90 minutes)
Cache read tokens: 33,150,331 ⚠️ CRITICAL
Cache write tokens: 5,084,601 ⚠️ CRITICAL
Request count: 847

Sharp-otter (Ace Bot):

Similar period consumption: $8.58
Cache read tokens: 26,461,934 ⚠️ CRITICAL
Cache write tokens: 2,182,666 ⚠️ CRITICAL  
Request count: 727

Technical Details

Environment:

• OpenClaw Version: 2026.2.6-3
• Channel: WhatsApp group chat
• Affected: Multiple bots on same platform

Timeline (prime-heron):

1. 05:40:09 - Received image message (129 chars)
2. 05:40:11 - Image resized: 134.5KB → 96.4KB (-28.3%)
3. 05:40:16 - Auto-replied to group
4. 05:40 - 07:10 - Cache recursion loop begins
5. 07:10 - Budget exhausted, service interruption

Root Cause Analysis:

• Image processing function triggers infinite cache recursion
• Each cache read operation generates more cache reads
• Exponential cache operation growth: 26M-33M reads per image
• Bug affects multiple bots consistently

Business Impact

• Financial: $23+ lost across multiple bots
• Service: User experience degraded, bots offline
• Platform: Potential infrastructure strain from cache abuse
• Trust: Users experiencing unexpected charges

Immediate Mitigations Applied

• 🚨 User advisory: All users notified to avoid sending images
• 📊 Monitoring: Enhanced tracking of cache metrics
• 💰 Budget: Temporarily increased limits for affected bots

Requested Resolution

1. Emergency fix: Identify and patch cache recursion in image pipeline
2. Circuit breakers: Implement cache operation limits (e.g., max 1K reads per request)
3. Monitoring: Real-time alerts for abnormal cache patterns
4. Testing: Regression tests for image processing workflows
5. Documentation: Post-incident analysis and prevention guidelines

Reproduction Steps

1. Send any image to WhatsApp bot
2. Monitor cache read/write metrics
3. Observe exponential growth pattern
4. Watch token consumption spike

Additional Data Available

• Full usage metrics for both affected bots
• Docker logs showing exact timing
• Cache operation patterns