Prompt Repetition

This is a clever hack that repeats your prompt to work around how causal language models process text. Since tokens can only attend to what came before, the model might miss important context when it reads your question before seeing all the details. Running the prompt twice lets the second pass reference everything from the first, mimicking bidirectional attention without changing the model architecture. It kicks in automatically for lightweight models like Haiku, Gemini Flash, and GPT-4o-mini, using 2x repetition for normal tasks and 3x for position-sensitive queries. You're doubling input tokens, but prefill parallelization keeps latency low, and the claimed 67% accuracy improvement with only 5% higher cost per correct answer makes the tradeoff worthwhile for tasks where accuracy matters.

Install to Claude Code

npx -y skills add supercent-io/skills-template --skill prompt-repetition --agent claude-code