Anthropic releases official prompting guide for Claude Opus 5: key recommendations

Anthropic has published an official guide on prompting for its Claude Opus 5 language model. In the instructions, the developers emphasize proper query configuration to help avoid excessive token consumption.

Requests for brevity and subagent control

The main difference between Opus 5 and previous versions is its tendency to provide detailed commentary on its own actions and automatically trigger subagents to complete tasks.

To prevent the model from exhausting its token limit on reasoning, Anthropic recommends:

• Adding direct instructions for brevity and conciseness to your prompts.
• Limiting deep process detailing when it is not strictly necessary.

Eliminating redundant instructions

The developers specifically note that requests such as “double-check your work” or “ensure the answer is correct” should be removed from prompts.

In Claude Opus 5, a self-verification algorithm is built-in by default—the model automatically verifies results during generation, making additional instructions in the prompt text redundant.

Where to find the guide

The full version of the prompt engineering guide is available in the official Anthropic documentation.