How-toMulti-sourceSelected73°

Opus 5.5 Official Prompt Guide Covers Effort Settings and Agent Behaviour

Why we picked it

The default shift from high to medium effort means existing prompts tuned for Opus 5 may behave differently, and some prompt blocks should now be deleted rather than kept. Builders running multi-agent or long-running workflows can apply the guide's concrete fixes, such as instructing the model to scan all relevant context before acting and telling it the remaining time budget to pace its work.

Anthropic has released an official prompt guide for Opus 5.

Body

5, which changes the default effort setting to medium.The company says Opus 5.

5 at medium matches or exceeds Opus 5 at high effort in its tests.Opus 5.

5 generates output tokens over 30% faster than Opus 5 and typically uses fewer tokens for the same task.The guide also explains that mid-task statements like 'next I plan to do C' are progress reports, not completion signals, and that real-time status output is hidden by default behind a dedicated toggle.

Original · 小互

Opus 5.5 官方提示词指南: effort 怎么设、Agent 为何中途停、哪些提示词该删? Opus 5.5 中 effort 的默认值改为了medium,Anthropic 称测试中 5.5 的 medium 达到或超过 Opus 5 的 high... Opus 5.5 生成输出 token 的速度比 Opus 5 快 30% 以上;完成同样的任务,通常用的 token 也更少 Opus 5.5在干活过程中和干完之后,汇报"我做了啥、发现了啥、还需要你干啥"讲得更清楚 所以: ⏸️ 长任务干到一半冒出一句「接下来我打算做 C」——这是在汇报进度,不是干完了,别误判掐断 🤫 干活间隙其实一直在小声汇报进展,默认看不见,得专门开一个开关才能看到实时状态 🔍 跨多个软件协作时容易漏看"没被点名但相关"的信息,提前叮嘱它"先都扫一遍再动手",准确率明显提升 ⏱️ 多个AI一起干活时,告诉它"还剩多少时间",它会自己调配节奏,干得更快 🎨 让它做网页设计,光说"别做得像AI"没用,得具体点名"别用米色背景/斜体标题/01-02-03编号"才管用📊 1 ⚡

Industry Coordinates & Source Verification

This report is categorized under AITOP's "How-to" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and AI regulation, compliance frameworks, and sovereign compute policy. Key entities and technical topics under continuous tracking include Opus 5.5, Anthropic, prompt engineering, agents.

Source attribution: curated from 小互 (@imxiaohu). Cross-source corroboration: 11 independent sources have covered this development (including shao__meng, lmarena.ai, elvis), confirming industry-wide signal strength.