RLM Prompt Example

Analyze the CSI (Customer Technical Support) Jira export at Jira-2.csv.gz (~709K rows, gzipped CSV) to identify recurring error patterns, top root causes, and systemic issues.

Use the RLM pattern: partition the data across analyst agents, analyze in parallel, synthesize.

> Note: The RLM pattern now auto-detects content types. For this CSV file, it will detect structured_data and route to swarm:rlm-data-analyzer agents with header-preserving chunks. See swarm:rlm-pattern for content-type detection and routing details.

1. Team Lead: Decompress and assess the dataset. Determine column structure and row count. Partition into 8-10 roughly equal CSV chunks (preserving the header row in each). Use Grep to scout for high-density error regions — prioritize partitions with the most incident volume.

2. Each of 8-10 Analyst agents: Read your assigned partition. For each chunk, report:

   • Top issue types and categories ranked by frequency
   • Recurring error patterns in summary/description fields (common phrases, failure signatures, repeated symptoms)
   • Component and product area breakdown
   • Priority and severity distributions
   • Resolution time statistics (min, median, p90, max) if resolution dates are available
   • Escalation and reassignment rates
   • Any outliers or anomalies worth highlighting

3. Team Lead: Collect all analyst reports and synthesize into a consolidated analysis:

   • Executive summary — total volume, date range, overall health assessment
   • Top 10 root causes ranked by frequency and impact, with representative ticket examples
   • Systemic issues — recurring patterns that indicate process, tooling, or architectural problems
   • Temporal trends — month-over-month or quarter-over-quarter changes in volume, categories, or severity
   • Component hotspots — which products, services, or modules generate the most support load
   • Recommendations — prioritized action items to reduce ticket volume, grouped by effort level (quick wins vs. structural fixes)