Best practices for agent design, token efficiency, and backlog prioritisation use cases

[p-nicolaou]

Hi team

First off — really enjoying working with Oh My OpenAgent. I’m exploring how to use agents more effectively in production workflows and had a few questions around best practices, especially when scaling usage.

Best practices for designing agents

• Are there recommended patterns for structuring agents (e.g. single-purpose vs multi-tool agents)?
• How do you typically decide when to:
  • Split logic across multiple agents
  • Keep everything within one agent with tools
  • Any guidance on orchestration vs delegation between agents?

Token efficiency becomes critical at scale, so I’m curious

• Are there recommended strategies for:
  • Reducing prompt size without losing performance?
  • Managing conversation memory effectively?

• Should we prefer:
  • Short system prompts + external context injection?
  • Or richer, self-contained prompts?
  • Any built-in features or patterns in this repo that help control token usage?

Use case: Analysing & prioritising a Jira backlog
I’m particularly interested in using agents for product/engineering workflows.
Example scenario:
• Input: A Jira backlog with ~100 tickets
• Each ticket includes:
• Title
• Description (some not clear)
• Labels

Goal:
• Cluster similar tickets
• Identify high-impact vs low-impact work
• Suggest prioritisation order
• Highlight quick wins vs complex tasks

Would you recommend:
• A single agent handling the full pipeline?
• Or multiple specialised agents (e.g. classifier → scorer → prioritiser)?
Any examples or patterns for:
• Batch processing large datasets like this?
• Avoiding excessive token usage when analysing many items?