Anthropic’s prompting guide for Claude Opus 5.5

Anthropic published a prompting guide for Claude Opus 5.5 covering effort calibration, thinking behaviour, unattended and multiagent runs, safeguard refusals, progress updates, multi-app workflows, pasted-text handling, and visual inputs. Opus 5.5 generates output tokens more than 30% faster than Opus 5 and often finishes the same task with fewer tokens; existing Opus 5 prompts should still work as a starting point. The doc maps common failure modes to the section to read first.