[Bug/Question] Excessive Input Token Usage When Invoking Skills - Rate Limit Errors

[alfredcs]

Affected Notebook/File

03_skills_custom_development.ipynb

Bug Description

When invoking skills/tools through the Claude agent, the input token consumption appears to be excessively high, causing frequent rate limit errors even for relatively simple operations.

RateLimitError: Error code: 429 - {
  'type': 'error', 
  'error': {
    'type': 'rate_limit_error', 
    'message': 'This request would exceed the rate limit for your organization of 50,000 input tokens per minute. Please reduce the prompt length or the maximum tokens requested, or try again later.'
  }, 
  'request_id': 'req_011CVwgfm56xSnTkeud2yzZX'
}

I'd like to understand and discuss potential strategies to reduce input token consumption:

1. System Prompt Optimization

• Is there a recommended way to minimize system prompt size when using agent skills?
• Can skill definitions be loaded dynamically rather than included in every request?

2. Conversation History Management

• What's the recommended approach for truncating or summarizing conversation history?
  Is there a sliding window implementation available?

3. Tool/Skill Definition Efficiency

• Are there best practices for defining tools with minimal token overhead?
• Can tool schemas be compressed or cached?

4. Caching Mechanisms

• Does the API support prompt caching to reduce repeated token charges?
• Are there plans to implement context caching for agent interactions?

5. Token Budgeting

• Is there a way to set a maximum input token budget per request?
• Can we get token count estimates before sending requests?

Steps to Reproduce

Error appear on : Test Brand Guidelines with Document Creation
Let's test the brand skill by creating a branded PowerPoint presentation:

Error Message

RateLimitError: Error code: 429 - {
  'type': 'error', 
  'error': {
    'type': 'rate_limit_error', 
    'message': 'This request would exceed the rate limit for your organization of 50,000 input tokens per minute. Please reduce the prompt length or the maximum tokens requested, or try again later.'
  }, 
  'request_id': 'req_011CVwgfm56xSnTkeud2yzZX'
}

Environment

No response

Would you be willing to submit a PR to fix this?

None