cost-limit — Token Budget Enforcement

Severity: CRITICAL · Auto-fix: No · Category: 💰 Cost

What It Does

Fires at CRITICAL level when the prompt exceeds the configured token_limit. Use this to catch accidentally bloated prompts before they hit API rate limits or blow your context window.

Example

Prompt: 850 tokens, token_limit: 800

[ CRITICAL ] cost-limit (line -)
  Prompt exceeds token limit: 850/800 tokens.
  Consider removing redundant content or splitting into smaller prompts.

Configuration

token_limit: 800  # Default

rules:
  cost_limit: true

Choosing a Limit

Set token_limit to 60–70% of your model's context window to leave room for the completion:

Model	Context window	Recommended token_limit
GPT-4o	128k	500–2000 (system prompt budget)
GPT-4	8k	500–1000
GPT-3.5 Turbo	16k	500–1500
Claude 3.5 Sonnet	200k	500–5000
Gemini 1.5 Pro	1M	500–10000

Disabling

rules:
  cost_limit: false

Or raise the limit:

token_limit: 2000