Delete every "verify your work" instruction from your Opus 5 prompts — Anthropic says they now cause over-verification with no quality gain

Claude Platform Docs·high signal

Anthropic's official Opus 5 prompting guide is explicit: if your prompt or harness contains instructions like "include a final verification step for any non-trivial task" or "use a subagent to verify," remove them, because Opus 5 verifies its own work and these instructions "cause over-verification" while "removing them reduces wasted tokens with no loss in quality." The same applies to "double-check your answer" and "re-verify before responding," which compound with the model's own self-correction. For anyone running a multi-agent pipeline, this is a direct token refund: the writer-verifier scaffolding built for Opus 4.x is now paying twice for behavior the model does for free.