Share you finding and workflow

[lst97]

🚀 Discussion: Claude Code Sub-Agent Document Feedback & Enhancements 🚀

I am opening up a discussion to gather your valuable feedback on the Claude Code Sub-Agent document! Our goal is to collectively identify areas for improvement, optimize performance, and explore new possibilities for our agentic workflows.

Please use this discussion to share:

1. Feedback on Existing Sub-Agent Documentation:

• Performance Issues:

  • Excessive Token Usage: Have you encountered instances where the sub-agent consumes an unexpectedly high number of tokens for a given task? Please provide specific examples if possible (e.g., input, expected output, observed token usage).
  • Execution Time / Long Runs: Are there scenarios where the sub-agent takes an unacceptably long time to complete its task? Screenshots of performance metrics (timestamps, duration) would be incredibly helpful.
  • Recursion / Looping Issues: Have you observed the agent getting stuck in recursive calls or infinite loops? Describe the triggers and provide any relevant logs or traces.
  • Unexpected Behavior/Errors: Any other bugs, inconsistencies, or deviations from expected behavior.

• Clarity and Completeness:

  • Is the documentation clear and easy to understand?
  • Are there any sections that are confusing or lack sufficient detail?
  • Are there missing examples or use cases that would improve comprehension?

2. New Ideas & Suggestions:

• New Agent Ideas: Do you have concepts for new sub-agents that could enhance our capabilities (e.g., specialized agents for specific coding tasks, debugging, testing)?
• Rule Enhancements: Suggestions for refining or adding new rules to improve agent decision-making, efficiency, or output quality.
• Tool Use Suggestions: Ideas for integrating new tools or improving the way existing tools are utilized by the sub-agents (e.g., linters, debuggers, static analysis tools, external APIs).

3. Performance Metrics, Usage Examples & Screenshots:

• Please share screenshots or detailed descriptions of the performance of the sub-agent usage. This could include:
  • Token usage graphs/logs.
  • Execution time breakdowns.
  • CPU/memory consumption (if applicable and observable).
  • Any other metrics that illustrate performance bottlenecks or successes.
• Feel free to share examples of how you're currently using the sub-agents and the actual outcomes you're observing, even if you don't have a specific suggestion for improvement. This helps us understand real-world application and performance.

To help us analyze your feedback effectively, please try to include:

• Context: Briefly describe the task or scenario where you observed the issue or have a suggestion.
• Reproduction Steps (if applicable): How can we reproduce the issue you encountered?
• Expected vs. Actual Behavior: What did you expect the sub-agent to do, and what did it actually do?
• Relevant Code Snippets or Configuration: If the issue is related to specific code or configuration, please include it.

I believe that by working together, we can significantly improve the efficiency, reliability, and utility of our Claude Code Sub-Agents.

Thank you for your contributions!

[arcofs]

Thanks for this, just started using. I am finding though that the orchestrator agent, regular stops. So it delegates the tasks to the sub agents, the sub agents do their thing, but I assume the sub agents dont report back to the orchestrator, and it stops there. For example:

`I need you to analyse the jobs and quotes funtionality of this system. In particular I want to be able to track costs on jobs, ongoing, budgets and costs at the end. You can see f we convert from a
quote we then have a budget to work with, but theres some things I want to discuss to see how the current system works, and whats possible and what needs doing. Some notes:

• We have machinery which can be assigned to a job in the @app\scheduling\timeline.tsx
• We also can assign technicins to jobs as well.
• Both machinery and techncicians have a cost rate and a chargeput rate
• Technicinas can create time sheets and allocate to jobs, this can hellp manage costs, but we dont have functionaoity to use time sheets for machinery. As machinery is not a person it may be labourius
  to create a time sheet for each machine for ech day by an admin or someone

Using the postgres tool look in the db
I want to understand what are the ways we currently track costs (if at all, and is it actually implemented, please check), and what are some ways we can easily track costs for suggestions on how to do
this better and more efficitenly . This goes for Materials, Machinery and Labour itself. Please look through the codebase and then come back to me to discuss
⎿  Read app\scheduling\timeline.tsx (2000+ lines)

● I'll analyze the jobs and quotes functionality to understand cost tracking capabilities. Let me start by examining the database structure and then explore the codebase.

● agent-organizer(Analyze cost tracking system)
⎿  Done (2 tool uses · 21.9k tokens · 18.5s)

● The agent-organizer is now analyzing your cost tracking system comprehensively. They'll examine the database structure, search through your codebase for existing cost implementations, and provide
detailed findings about how materials, machinery, and labor costs are currently handled, along with specific recommendations for improvements.`