Sentry's Span Hierarchy Exposed a Silent Retry in My 5-Agent Pipeline. One Agent Took 22.6s, the Others Took 5.

Short summary

A developer built a 5-agent AWS security scanning pipeline on CrewAI with Bedrock Nova Pro and used Sentry's span hierarchy to diagnose why one agent took 22.6s while others took 5-10s. The root cause was an IAMAnalyzer tool dumping a 27KB JSON blob into the LLM context, triggering silent retries. The fix involved paginating results and limiting to the 20 most active roles, cutting scan time dramatically.

• •5-agent CrewAI pipeline on Bedrock Nova Pro scans AWS accounts for security misconfigurations
• •Sentry trace waterfall revealed a 27KB context payload causing silent LLM retries in the SecurityScanner agent
• •Fix: paginate IAM role results and limit to top 20 most active roles, reducing agent time from 22.6s to normal range

Generated with AI, which can make mistakes.