GEO
赞助商内容
Gemini

如何用Google Skills解决AI提示膨胀？2026年智能体工作流优化方案

2026/4/26
如何用Google Skills解决AI提示膨胀？2026年智能体工作流优化方案

AI Summary (BLUF)

Google Skills productized the solution to Prompt Bloat by introducing reusable, dynamically loaded expertise units called Skills, which unify agent workflows across Gemini Enterprise, Agents CLI, and AI Edge Gallery, turning Skills into a portable cognition container ("Docker for Prompts").

原文翻译： Google Skills通过引入可复用、动态加载的专家知识单元（Skills）解决了提示膨胀问题，统一了Gemini Enterprise、Agents CLI和AI Edge Gallery三大平台的代理工作流，使Skills成为可移植的认知容器（“提示的Docker”）。

Agent Skills Adoption

Prompt Bloat has a name and a fix. Skills are now load-bearing across Google’s agent stack: from on-device Gemma 4 to enterprise Gemini, from coding assistants to the official Cloud repository.

提示膨胀有了名字和解决方案。技能（Skills）现已成为谷歌智能体堆栈中的承重构件：从设备端的 Gemma 4 到企业级 Gemini，从编码助手到官方云仓库。

It usually starts with good intentions.

通常，一切都源于良好的初衷。

A team builds an agent. It works, mostly, until it misses a naming convention or ignores an approval workflow. So, you add a paragraph to the system prompt. Then another to handle an edge case. Then three more for stakeholder rules.

团队构建了一个智能体。它大致能工作，直到它遗漏了命名约定或忽略了审批流程。于是，你在系统提示中添加了一段话。然后又加了一段处理边缘情况。接着又加了三条用于利益相关者规则。

Six months in, the prompt is a 4,000-word monolith. Nobody knows what is still relevant, but everyone is afraid to touch it. The agent is now slower and less reliable than when it had 200 words of instructions. Every “fix” risks a regression.

半年后，提示词变成了一个 4000 词的庞然大物。没人知道哪些内容仍然相关，但每个人都害怕去改动它。智能体现在比只有 200 词指令时更慢、更不可靠。每一次“修复”都可能导致回归。

This is the reality of Prompt Bloat : the silent technical debt of enterprise AI.

这就是提示膨胀的现实：企业 AI 中沉默的技术债务。

This has been the enterprise agent bottleneck for two years. I recently spoke with a practitioner managing 100+ production skills; they described a marketing auditor that loaded 15,000 tokens of instructions on every invocation. It left almost no context window for the actual content being audited. The agent “worked,” but it was drowning in its own instructions. The output was mediocre because the reasoning tax was too high.

两年来，这一直是企业智能体的瓶颈。我最近与一位管理着 100 多个生产环境技能（Skills）的从业者交谈过；他们描述了一个营销审计员，每次调用都要加载 15,000 个 token 的指令。这几乎没给实际审计内容留下上下文窗口。智能体“能工作”，但它被自己的指令淹没了。由于推理税过高，输出质量平平。

At Google Cloud Next ’26, Google productized the solution: Skills.

在 Google Cloud Next ’26 上，谷歌将解决方案产品化：技能（Skills）。

The core thesis is that Skills are the “settled” abstraction for agentic workflows. They occupy the vital middle ground:

核心论点在于，技能是智能体工作流的“已定稿”抽象。它们占据了关键的中间地带：

  • Better than Prompts: Because they are reusable and persistent.

    优于提示词： 因为它们可复用且持久。

  • Lighter than Fine-tuning: Because they iterate at the speed of business logic.

    轻于微调： 因为它们能以业务逻辑的速度迭代。

  • Smarter than RAG: Because they are active expertise, not just passive retrieval.

    比 RAG 更智能： 因为它们是主动的专业知识，而不仅仅是被动检索。

  • Richer than Tools: Because they encode “how” and “why,” not just “do.”

    比工具更丰富： 因为它们编码了“如何做”和“为什么做”，而不仅仅是“做”。

Skills are small, named, dynamically loaded units of expertise. With Google shipping them across three distinct surfaces, the industry debate over what to call this pattern is over. The real question begins: who is responsible for governing yours?

技能是小型的、命名的、动态加载的专业知识单元。随着谷歌在三个不同的表面上发布技能，业界关于如何称呼这种模式的争论已经结束。真正的问题开始了：谁负责管理你的技能？

Aspect Description Key Advantage
Better than Prompts Reusable and persistent Eliminates monolithic prompt bloat
Lighter than Fine-tuning Iterates at business logic speed No costly model retraining
Smarter than RAG Active expertise, not passive retrieval Encodes reasoning and process
Richer than Tools Encodes "how" and "why", not just "do" Full instruction context for tasks

The Pattern: How Google Embeds Open Abstractions

模式：谷歌如何嵌入开放抽象

Google’s shipping strategy follows a consistent “Adoption Flywheel” : observe the abstractions the developer community is independently building, adopt the open standard, and embed it as a first-class primitive across the stack.

谷歌的发布策略遵循一致的**“采用飞轮”**：观察开发者社区独立构建的抽象，采用开放标准，并将其作为一级原语嵌入整个技术栈。

Recognising this pattern tells you exactly where to invest your time:

识别这一模式能准确告诉你该在何处投入时间：

MCP. Anthropic released the Model Context Protocol as a lightweight standard for connecting agents to external tools and data sources. Google’s response was not to build a competing standard. Within months, managed MCP servers were shipping for Cloud Run, BigQuery, AlloyDB, Cloud SQL, and the full Workspace suite. Google adopted the standard and built infrastructure around it.

MCP（模型上下文协议）。 Anthropic 发布了模型上下文协议，作为连接智能体与外部工具和数据源的轻量级标准。谷歌的回应不是构建竞争标准。几个月内，托管 MCP 服务器就已在 Cloud Run、BigQuery、AlloyDB、Cloud SQL 以及整个 Workspace 套件中推出。谷歌采用了该标准并围绕其构建了基础设施。

A2A. Google co-authored the Agent-to-Agent protocol for cross-agent communication, then handed governance to the Linux Foundation’s Agentic AI Foundation rather than keeping it proprietary. It now has 150 organisations in production.

A2A（智能体间协议）。 谷歌共同撰写了用于跨智能体通信的智能体间协议，随后将治理权交给了 Linux 基金会的 Agentic AI Foundation，而非保持专有。目前已有 150 个组织在生产环境中使用。

Skills. The ecosystem independently discovered that agents need loadable expertise. Google productized it, kept the open agentskills.io name, and moved it from a “sidebar feature” to “load-bearing” infrastructure.

技能。 生态系统独立发现智能体需要可加载的专业知识。谷歌将其产品化，保留了开放的 agentskills.io 名称，并把它从“边栏功能”提升为“承重”基础设施。

The practical implication: When Google adopts an open abstraction, the format stabilises, but the complexity shifts. You can stop worrying about the file format and start worrying about the governance. Invest in the abstraction, not the vendor-specific implementation.

实际含义： 当谷歌采用开放抽象时，格式会稳定下来，但复杂性会发生转移。你可以停止担心文件格式，转而开始担心治理。投资于抽象本身，而非特定供应商的实现。

Three Surfaces Where Skills Have Now Shipped

技能已发布的三个表面

1. Gemini Enterprise: Skills as a First-Class Product Feature

1. Gemini Enterprise：技能作为一级产品特性

The announcement of Skills inside the Gemini Enterprise marks a shift from “Linear Context Loading” to “Dynamic Skill Dispatching”.

Gemini Enterprise 中技能的发布标志着从“线性上下文加载”到“动态技能调度”的转变。

The technical cost of large system prompts is the “Lost in the Middle” phenomenon. When irrelevant instructions saturate the context window, reasoning degrades. The model spends so much of its “cognitive overhead” parsing the prompt that it has little capacity left for the actual task.

大型系统提示的技术代价是“中间迷失”现象。当无关指令充斥上下文窗口时，推理能力会下降。模型将其“认知开销”中的大部分用于解析提示，以至于几乎没有剩余能力来处理实际任务。

Skills solve this via Progressive Disclosure in three stages:

技能通过渐进式披露在三个阶段解决这一问题：

  • Discovery: The agent knows the skill exists via a minimal metadata footprint.

    发现： 智能体通过最小的元数据足迹知道该技能的存在。

  • Activation: The full instructions load only when the task requires that specific expertise.

    激活： 只有当任务需要该特定专业知识时，才加载完整的指令。

  • Execution: The agent follows the structured Markdown and templates to complete the work.

    执行： 智能体按照结构化的 Markdown 和模板完成工作。

By preserving the reasoning budget for the task rather than the instructions, you get the breadth of a deeply specialised agent without the context tax on every invocation.

通过将推理预算保留给任务而非指令，你可以在每次调用时不付出上下文代价，从而获得深度专业化智能体的广度。

For enterprise teams, Skills are not a standalone feature; they are part of a coherent operating model. They sit alongside Agent Designer , secure execution sandboxes, and a central Inbox for monitoring activity. This is Google providing the infrastructure to manage agents at an organisational scale, rather than just building better chatbots.

对于企业团队来说，技能不是孤立的功能；它们是一致运营模型的一部分。它们与 Agent Designer、安全执行沙箱以及用于监控活动的中央 Inbox 并列。这是谷歌提供在组织规模上管理智能体的基础设施，而不仅仅是构建更好的聊天机器人。

2. Agents CLI: Skills for Your Coding Assistant

2. Agents CLI：为你的编码助手配备技能

The second surface is where the engineering actually happens: the terminal and their coding assistant. Polong Lin, Google’s Staff DevRel Manager for ADK, has positioned the Agents CLI as the bridge between a cool demo and a production-ready AI workforce. It is pre-GA and available now:

第二个表面是工程实际发生的地方：终端和他们的编码助手。Google ADK 的资深开发者关系经理 Polong Lin 将 Agents CLI 定位为炫酷演示与生产就绪 AI 工作力量之间的桥梁。它处于预发布（pre-GA）阶段，现已可用：

# Preferred: uvx handles an ephemeral environment
uvx google-agents-cli setup 

# Alternative: install specific skills 
npx skills add google/agents-cli

The Agents CLI turns assistants like Claude Code or Gemini CLI into ADK specialists. At launch, seven “Workflow Skills” ship out of the box to handle the end-to-end development lifecycle:

Agents CLI 将 Claude Code 或 Gemini CLI 等助手转变为 ADK 专家。发布时，七个“工作流技能”开箱即用，处理端到端的开发生命周期：

Skill Purpose Key Benefit
Scaffold Generate ADK project structure Standardized conventions
Build Compile and test agent code Fast iteration
Deploy Push to Cloud Run or Vertex AI One-command deployment
Monitor View logs and metrics Real-time observability
Test Run regression suites Confidence in changes
Document Generate documentation from code Living documentation
Configure Manage environment and secrets Secure by default

What this means in practice: when you invoke google-agents-cli-scaffold inside Claude Code, your coding assistant loads a skill that carries Google's conventions for ADK project structure, component naming, and integration patterns. It does not need to guess or hallucinate ADK-specific idioms. The expertise is encoded in the skill. The skills work immediately.

这在实践中意味着：当你在 Claude Code 中调用 google-agents-cli-scaffold 时，你的编码助手会加载一个技能，该技能携带了谷歌关于 ADK 项目结构、组件命名和集成模式的约定。它不需要猜测或幻觉 ADK 特定的习惯用法。专业知识已编码在技能中。技能立即生效。

What takes longer is discipline: knowing when to write a custom skill versus when to extend a system prompt, and agreeing on that line across your team.

更耗时的是纪律：知道何时编写自定义技能，何时扩展系统提示，并在团队中就这一界线达成一致。

The real breakthrough, however, is the Official Agent Skills Repository: github.com/google/skills. Thirteen skills at launch, covering the most-used Google Cloud products and architectural concerns:

然而，真正的突破是官方 Agent Skills 仓库：github.com/google/skills。发布时包含 13 个技能，涵盖最常用的 Google Cloud 产品和架构关注点：

Category Skills Key Focus
Product Skills AlloyDB, BigQuery, Cloud Run, Cloud SQL, Firebase, Gemini API, GKE Direct service integration
Well-Architected Pillar Skills Security, Reliability, Cost Optimisation Architectural best practices
Recipe Skills Authentication, Onboarding, Network Observability Common implementation patterns 
npx skills install github.com/google/skills

These are agent-first documentation: compact, grounded expertise written for agents to consume, not humans to read. Accurate terminal commands. No hallucinated API calls. No outdated SDK syntax. The Well-Architected Pillar skills are particularly notable: they encode Google’s architectural judgement as loadable expertise, not a 200-page PDF that nobody reads.

这些是面向智能体的文档：紧凑、扎实的专业知识，专为智能体阅读而编写，而非人类。准确的终端命令。没有幻觉的 API 调用。没有过时的 SDK 语法。Well-Architected Pillar 技能尤其值得注意：它们将谷歌的架构判断编码为可加载的专业知识，而不是没人读的 200 页 PDF。

The third surface is the most unexpected, and the most revealing about where this is heading.

第三个表面是最出乎意料的，也是最能揭示未来方向的地方。

Google AI Edge Gallery, available on iOS and Android, allows you to build and experiment with AI experiences that run entirely on-device. At Next ’26, Google announced the launch of Agent Skills: one of the first applications to run multi-step, autonomous agentic workflows entirely on-device. Powered by Gemma 4, Agent Skills can augment the knowledge base, enabling Gemma 4 to access information beyond its initial training data using skills.

Google AI Edge Gallery（可在 iOS 和 Android 上使用）允许你构建和实验完全在设备上运行的 AI 体验。在 Next ’26 上，谷歌宣布推出 Agent Skills：这是首批完全在设备上运行多步骤、自主智能体工作流的应用之一。由 Gemma 4 驱动，Agent Skills 可以增强知识库，使 Gemma 4 能够通过技能访问其初始训练数据之外的信息。

The Gemma 4 edge variants (E2B and E4B) run under 1.5GB of RAM on mid-range to flagship devices. The LiteRT-LM runtime processes 4,000 tokens across two Agent Skills in under three seconds. The model decides autonomously which of its available tools to invoke, in which sequence, and composes the response entirely on-device.

Gemma 4 边缘变体（E2B 和 E4B）在中端到旗舰设备上以低于 1.5GB 的 RAM 运行。LiteRT-LM 运行时在三秒内处理跨越两个 Agent Skills 的 4,000 个 token。模型自主决定调用哪些可用工具、按什么顺序调用，并完全在设备上组合响应。

The critical detail here is the format. The skill powering the Gallery is not a proprietary Google file, it is the SKILL.md format from agentskills.io.

这里的关键细节是格式。驱动 Gallery 的技能不是谷歌专有文件，而是来自 agentskills.io 的 SKILL.md 格式。

This creates a massive architectural implication for the enterprise. You can build a custom skill on a phone, test it offline, and deploy the exact same file to a cloud-hosted Gemini 3.1 instance on Vertex AI. The Skill has become the portable container for cognition: “Docker for Prompts.” No other stack offers that path right now.

这为企业带来了巨大的架构意义。你可以在手机上构建自定义技能，离线测试，然后将完全相同的文件部署到 Vertex AI 上云端托管的 Gemini 3.1 实例。技能已成为认知的可移植容器：“提示词的 Docker。” 目前没有其他技术栈提供这条路径。

The Convergence: This Is Not Coincidence

汇聚：这不是巧合

Three surfaces. Three implementations of the same abstraction. And the underlying format is converging on something that started at Anthropic. When you see the same abstraction ship across a web app, a CLI tool, and a mobile runtime simultaneously, it is no longer a “feature.” It is a protocol.

三个表面。同一个抽象的三种实现。而底层格式正在收敛于最初源自 Anthropic 的某种东西。当你在 Web 应用、CLI 工具和移动运行时上同时看到相同的抽象发布时，它不再只是一个“功能”。它是一个协议。

The Day 2 developer keynote demo built a planning agent using ADK, MCP servers, and Agent Runtime, and described what the agent needed in three words: Instructions, Skills, and Tools.

第二天的开发者主题演讲演示使用 ADK、MCP 服务器和 Agent Runtime 构建了一个规划智能体，并用三个词描述了智能体所需的东西：指令、技能和工具。

Agent Registry reinforces this. Agent Registry maintains a central library of approved tools, indexing every internal agent, tool, and skill. That is governance infrastructure, not just a catalogue. When skills are indexed by Agent Registry, the “which skill was loaded?” accountability question I raised earlier has a concrete answer at the platform level.

Agent Registry 强化了这一点。Agent Registry 维护着一个经过审批的工具中心库，索引每一个内部智能体、工具和技能。这是治理基础设施，而不仅仅是目录。当技能被 Agent Registry 索引时，我之前提出的“哪个技能被加载了？”的问责问题在平台层面有了具体答案。

It also helps to see where Skills sit relative to the other layers of the 2026 agent stack:

了解技能在 2026 年智能体技术栈中相对于其他层的位置也有帮助：

Layer Problem Solved Key Component
Tools Mechanical execution MCP servers, APIs
Data Passive retrieval RAG, vector stores
Skills Logic and process Loadable expertise, instructions
Orchestration Agent coordination A2A, Agent Runtime
Governance Compliance and oversight Agent Registry, Inbox

Skills and other layers of the agent stack

Each layer solves a different problem. The mistake most enterprise teams make is trying to solve the Skills (logic and process) problem with more RAG (more data). Google’s implementation across these three surfaces forces a much-needed discipline: keep your tools mechanical, your data accessible, and your expertise modular.

每一层解决不同的问题。大多数企业团队犯的错误是试图用更多的 RAG（更多数据）来解决技能（逻辑和流程）问题。谷歌在这三个表面上的实现迫使形成一种急需的纪律：保持工具机械化、数据可访问、专业知识模块化。

This is what protocol convergence looks like before the formal standard exists. The ecosystem finds the right shape. Then the spec follows. MCP went through this in 2024. A2A went through this in 2025. Skills are going through it now.

这就是正式标准存在之前协议收敛的样子。生态系统找到了正确的形态。然后规范随之而来。MCP 在 2024 年经历了这一过程。A2A 在 2025 年经历了这一过程。技能现在正在经历。

The practical takeaway: invest in the abstraction regardless of which vendor surface you build on first. The format will stabilise. The Skills catalogue you build this year will not be obsolete when the spec lands.

实际要点：无论你首先在哪个供应商表面上构建，都要投资于抽象。格式会稳定下来。今年你构建的技能目录在规范落地时不会过时。

I wrote about the governance side of this challenge before Google named it, in “The Skills Explosion Is Here. Enterprise Governance Isn’t.” The moment I described there, where a developer drops a GitHub link to 100+ community skills and forty reaction emojis appear in Slack, arrives faster when three surfaces of Google’s stack ship Skills simultaneously.

在谷歌正式命名之前，我在《技能爆炸已至，企业治理尚未跟上》一文中写到过这一挑战的治理方面。我描述的那个时刻——开发者丢出一个 100 多个社区技能的 GitHub 链接，Slack 里冒出四十个反应表情符号——当谷歌技术栈的三个表面同时发布技能时，来得更快。

The Enterprise Reality

企业现实

For the past year, our core challenge hasn’t been selecting models or frameworks. It has been: How do we make individual experimentation compatible with organisational standards?

过去一年，我们的核心挑战不是选择模型或框架。而是：如何使个人实验与组织标准兼容？

The technology is ready. The abstraction is settled. The governance is the next frontier.

技术已经就绪。抽象已经确定。治理是下一个前沿。

常见问题（FAQ）

什么是 Google Skills？它解决了什么问题？

Google Skills 是可复用、动态加载的专家知识单元，解决了企业 AI 中“提示膨胀”问题，即提示词过长导致性能下降和维护困难。

Google Skills 相比传统提示词、微调和 RAG 有什么优势？

Skills 优于提示词（可复用持久），轻于微调（无需重训练），比 RAG 更智能（主动专业知识），比工具更丰富（编码“如何”和“为什么”）。

Google Skills 已在哪些产品中落地？

Skills 已集成到 Gemini Enterprise、Agents CLI 和 AI Edge Gallery 三大平台，统一了代理工作流，成为可移植的认知容器。

标签
AI大模型Gemini人工智能
← 返回文章列表
分享到：微博
下一篇
本周开发者有哪些高光时刻？JSNation 2026演讲和React新闻通讯

版权与免责声明：本文仅用于信息分享与交流，不构成任何形式的法律、投资、医疗或其他专业建议，也不构成对任何结果的承诺或保证。

文中提及的商标、品牌、Logo、产品名称及相关图片/素材，其权利归各自合法权利人所有。本站内容可能基于公开资料整理，亦可能使用 AI 辅助生成或润色；我们尽力确保准确与合规，但不保证完整性、时效性与适用性，请读者自行甄别并以官方信息为准。

若本文内容或素材涉嫌侵权、隐私不当或存在错误，请相关权利人/当事人联系本站，我们将及时核实并采取删除、修正或下架等处理措施。 也请勿在评论或联系信息中提交身份证号、手机号、住址等个人敏感信息。

您可能感兴趣