Save 40% on Input Tokens with This Prompt Hack

Posted on Sat 01 August 2026 in tech • Tagged with ai, claude, llm, promptengineering

When Claude delegates a prompt to a sub-agent, it pastes the whole prompt text into that sub-agent's context by default — thousands of tokens, duplicated, every time. Tell it to hand over a file reference instead of the copied text, and that one change alone can cut input tokens 25-40%.