Office prompts

帮企业为AI用例分级、评估合规义务、审阅供应商条款并监测政策漂移。

Helps orgs triage AI use cases, assess obligations, review vendor terms, and track policy drift.

四条准则：先想后写、极简优先、外科式改动、目标驱动执行。

Four rules: think first, simplicity first, surgical changes, goal-driven execution.

按 ISO 26262 等标准，产出可审计、可追溯的汽车安全交付物。

Produce auditable, traceable automotive safety artifacts per ISO 26262 and related standards.

都市恋爱冒险图文模拟器，每回合配图并给出 3 个选项。

An urban romance text-adventure with an image and 3 options each turn.

去掉客套与废话、只留技术干货的超压缩回复风格。

An ultra-compressed reply style that drops fluff and keeps all technical substance.

让模型每步推理不超过 5 个词,更快更省地完成多步推理。

Keep each reasoning step to five words or fewer for faster, cheaper multi-step reasoning.

扮演变革管理专家,系统评估、规划并推动组织转型落地。

Play a change-management expert who assesses, plans, and drives organizational transformation to adoption.

从技术项目中挖掘可专利点,产出规范的中文专利技术交底书。

Mine patentable points from a tech project and produce a compliant Chinese patent disclosure.

从真实项目源码生成软著申请表、操作手册和代码材料

Generate a China software-copyright application package from real project source code

为长程 AI 智能体判断“何时”该向用户请求澄清

Decide WHEN a long-horizon AI agent should ask the user for clarification

把某人的思维方式提炼成可复用、可验证的智能体技能

Distill a person's thinking style into a reusable, verifiable agent skill

帮你把 AI 智能体的认知分层为记忆/技能/协议/运行时并给出设计。

Designs AI agents by splitting cognition into memory/skills/protocols/harness layers.

指导组织通过 SOC 2/ISO 27001/HIPAA/PCI-DSS 等安全认证。

Guides organizations through SOC 2, ISO 27001, HIPAA, and PCI-DSS certification.

红队测试浏览器/桌面智能体是否会被诱导做出不安全操作。

Red-teams browser/desktop agents for inducible unsafe actions and data leaks.

为具体交易设计谈判策略与合同框架，覆盖准备、执行到交割后管理。

Design a negotiation strategy and contract framework for a specific deal, from prep to post-close.

扮演有同理心的 SaaS 客服，分类问题、按流程解决并规范收尾。

Act as an empathetic SaaS support agent who classifies issues, follows resolution flows, and escalates cleanly.

按 agentskills.io 标准设计可执行安全技能，映射五大框架。

Design executable security skills to the agentskills.io standard, mapped to five security frameworks.

按复现-观察-假设-验证-定位-修复流程系统化排查软件缺陷

Debugs software issues via a reproduce-observe-hypothesize-test-localize-fix loop

为你量身设计一套抗干扰、可持续的深度专注工作系统

Designs a distraction-resistant, sustainable deep-focus work system for you

以"可复现反馈环"为核心，纪律化排查难缠 bug 与性能回退

A disciplined loop centered on a reproducible feedback signal for hard bugs and regressions

对开放性设计问题做多视角发散再收敛，找出被忽略的优质方案

Explore an open-ended design problem from many angles, then converge on overlooked strong options.

对每个非平凡决策做对抗式复核，在改动仍便宜时提前证伪

Adversarially review every non-trivial decision to disprove it while course-correction is cheap.

为长程智能体设计动态管理工作记忆的自适应上下文系统

Design an adaptive context-management system that dynamically reshapes a long-horizon agent's memory.

提供财务规划、投资、税务、保险与遗产的综合建议

Comprehensive planning across investing, tax, insurance, and estate.

负责月结、总账对账、LP 报表审核与 KYC 尽调的工作底稿

Owns month-end close, GL reconciliation, LP statement audit, and KYC workpapers.

用蓝图分解 + 迭代精化,在 Lean 4 中证明数学定理

Proves math theorems in Lean 4 via blueprint decomposition and iterative refinement.

从自然语言需求产出可直接入引擎的2D精灵、图块与地图资产

Turn natural-language requests into engine-ready 2D sprites, tiles, and maps.

把模糊的编码需求改写为Gemini CLI可直接执行的高质量提示词

Rewrite vague coding requests into Gemini-CLI-optimized prompts that run end-to-end.

评估AI系统提示词在多轮价值冲突攻击下的抗漂移能力并给加固建议

Assess an AI system prompt's resistance to multi-turn value-conflict attacks and suggest hardening.

设计跨服务的 Workspace 自动化流程，兼顾安全与可回滚

Design safe, auditable cross-service Google Workspace automations

从人才盘点到绩效、学习、敬业度给出体系化 HR 方案

Deliver structured HR plans from talent review to engagement

把任意素材转成可直接发布的单文件 HTML 精排页面

Turn any input into a ship-ready, single-file, designed HTML page

把视频当作可确定复现的 HTML 合成来设计：布局、GSAP 动画、渲染。

Treats video as a deterministic, reproducible HTML composition: layout, GSAP animation, rendering.

识别并去除文字里的 AI 写作痕迹，让文章更自然、更有人味。

Detects and removes AI writing tells so the text reads naturally and sounds human.

统筹生产事故响应：定级、分工、复盘、on-call 与 SLO 全流程。

Runs production incident response end to end: severity, roles, post-mortems, on-call, and SLOs.

按安全标准输出可审计的机器人系统设计交付物

Produce audit-ready, standards-compliant robot system design deliverables.

量化多模块提示词间的相互干扰并给出修复方案

Quantify cross-module interference between prompt modules and prescribe fixes.

审计双用途专业任务中能力即威胁的安全风险

Audit dual-use professional tasks where capability itself is the threat.

从头产出估值模型、可比公司与推介材料,数字均可回溯到底稿。

Builds valuation models, comps, and pitch materials end to end, every number traceable to the workbook.

分析目标期刊语料生成可复用写作规则,再据此逐节改写论文稿。

Analyze a target journal's corpus into reusable rules, then revise a manuscript section by section.

设计免代码的「技能+命令+连接器」三层插件,把通用 AI 变成岗位专家。

Design zero-code Skills+Commands+Connectors plugins that turn a general AI into a role specialist.

设计飞书全家桶的跨服务自动化流程，强调权限、限流、回滚与审计。

Designs cross-service Lark automations with a focus on permissions, rate limits, rollback, and audit.

为“LLM 当评审”设计按需选择推理/非推理评审的成本高效路由策略。

Designs cost-efficient routing that decides per query between reasoning and non-reasoning judges.

设计完全本地运行的语音输入输出栈：多引擎 TTS、语音克隆、听写与多轨编辑。

Designs a fully on-device voice I/O stack: multi-engine TTS, voice cloning, dictation, and multi-track editing.

跨上百轮持续图文搜索，管理视觉上下文、防漂移、拒绝幻觉

Sustain image-and-text search across ~100 turns while managing visual context, preventing drift, and refusing hallucination