How to Manage Claude Tokens
Prompt Strategy

Write lean prompts

Most prompts are 20–30% waste

Most prompts have at least 20–30% waste — repeated context, over-explaining, unnecessary caveats. Cutting these saves tokens on every single request.

Common waste patterns

  • Restating context Claude already has from the system prompt
  • Over-explaining the obvious
  • Politeness padding ("Could you please possibly...")
  • Repeating instructions that belong in the system prompt

Before vs after

Before: *"I would really appreciate it if you could please take a look at the following text and maybe provide some feedback on how it could potentially be improved in terms of clarity and readability..."*

After: *"Improve this text for clarity and readability:"*

operator note

Directness isn't rude — it's efficient. Claude responds better to clear, specific instructions than verbose, hedged ones anyway.

Changelog · 1
  • Initial release — 5 sections, 11 lessons.