1000 万激励 + 100 位 AI 导演,LibTV 这一步把 agent 应用生态从"工具"卷到了"交易"

Alongside the official release of LibTV Agent, the platform unveiled what it describes as one of the world’s largest professional video Skill Hubs. Its first collection includes more than 100 director-grade video Skills covering film and television production, brand advertising, e-commerce marketing, short-form drama, music videos and creator content.

RMB 10 Million in Incentives and 100+ AI Directors: The Agent App Economy Is Moving From “Tools” to “Transactions”

Bilingual Feature|English & Chinese
English version first. The complete Chinese version begins after 中文版.

On July 14, 2026, LibTV made a move that may prove more important than the launch of another AI video feature.

Alongside the official release of LibTV Agent, the platform unveiled what it describes as one of the world’s largest professional video Skill Hubs. Its first collection includes more than 100 director-grade video Skills covering film and television production, brand advertising, e-commerce marketing, short-form drama, music videos and creator content.

Each Skill packages far more than a visual preset. It can contain:

  • A complete creative workflow;

  • Prompt and instruction strategies;

  • Model-selection logic;

  • Style-control methods;

  • Storyboard conventions;

  • Editing and pacing rules;

  • Industry-specific production experience.

Creators can invoke an existing Skill directly or turn their own production expertise into a new Skill that others can use.

LibTV then added a more aggressive incentive:

An RMB 10 million “Skill Super-Creation Incentive Plan.”

High-quality Skill creators may receive official traffic support, cash rewards, brand co-creation opportunities and potential long-term revenue sharing.

This is not merely the launch of a product feature.

It is the beginning of a marketplace.


From “Using an AI Tool” to “Buying a Capability”

The first generation of AI video creation placed most of the burden on the user.

To produce a usable video, creators had to:

  • Write detailed prompts;

  • Adjust generation parameters;

  • Choose among several models;

  • Test different visual styles;

  • Regenerate failed shots;

  • Manually correct pacing;

  • Reassemble clips;

  • Repeat the process until the result became acceptable.

The user was technically operating an AI tool, but in practice still had to perform much of the director’s work.

LibTV Agent proposes a different interaction model:

Choose a Skill, then let the Agent apply the complete methodology behind it.

Instead of describing every shot, creators can select a director-level capability.

A Wes Anderson-inspired Skill may understand symmetrical composition, precise staging and controlled color palettes.

A classical martial-arts Skill may apply spatial tension, wide-shot rhythm and traditional visual movement.

A Pixar-style advertising Skill may combine emotional storytelling, character-driven pacing and a brand-friendly final reveal.

The user no longer needs to teach the Agent how to direct every scene.

The Skill already contains the method.


A Skill Is Not a Template

The distinction between a Skill and a traditional template is essential.

A template is static.

Its parameters, structure and output logic are largely fixed. Users place their own content into an existing frame.

A Skill is dynamic.

It can package a chain of professional judgments:

  • When should the shot change?

  • Which emotion should use a warm or cold palette?

  • How quickly should the product appear?

  • When should the story slow down?

  • How should the Agent respond when reference materials conflict?

  • Which model should handle scripting, storyboarding or final review?

  • How should the ending reinforce brand recognition?

The Agent does not simply imitate a visual shell.

It applies a complete creative method to the context of the current project.

This changes the role of AI video creation.

Previously, users communicated with the model through increasingly detailed prompts:

Generate this type of video, using this lighting, this framing, this pacing and this editing style.

The longer the instructions became, the easier it was for the output to drift, contradict itself or lose consistency across scenes.

With a Skill, the user can communicate at a higher level:

Apply this director’s methodology.

The Agent handles the decision chain underneath.


Professional Experience Becomes a Transferable Asset

The real significance of a Skill Store is not that it offers more styles.

It converts professional expertise into a reusable and distributable asset.

In the traditional creative industry, a director’s most valuable knowledge lives inside personal experience:

  • How to establish a mood in the opening seconds;

  • When to reveal a product;

  • How to balance storytelling with conversion;

  • How to control rhythm without losing attention;

  • Which visual details create a premium feeling;

  • How to end on a memorable brand frame.

That judgment may have taken years and dozens of real projects to develop.

Yet it can normally serve only the limited number of productions the director personally joins.

Once that methodology is structured and packaged as a Skill, it can be invoked repeatedly by creators around the world.

The transformation looks like this:

Personal expertise
→ Structured methodology
→ Agent Skill
→ Marketplace distribution
→ Repeatable revenue

A director no longer sells only personal working hours.

The director may also distribute a digital version of professional judgment.

That is why a Skill is not merely a tool component.

It is a tradable capability.


The End of Prompt Engineering as the Main Interface

The rise of Skill Stores also suggests that the central interface of generative AI is changing.

The first phase of AI adoption was dominated by prompt engineering.

Users tried to discover the perfect wording, write longer instructions and build complex prompt templates.

The Agent era shifts the question.

Instead of asking:

How should I explain this job to the model?

Users increasingly ask:

Which capability should I install or invoke?

That is a far more scalable interaction model.

Most people do not want to learn professional directing terminology, complex editing logic or the technical differences among dozens of models.

They want to select a trusted capability and receive a usable result.

The same structure can extend beyond video creation:

  • Contract-review Skills;

  • Financial-analysis Skills;

  • Code-migration Skills;

  • Customer-service Skills;

  • Cybersecurity-response Skills;

  • Medical-document Skills;

  • Sales-follow-up Skills;

  • Cross-border marketing Skills.

Prompts describe the objective.

Skills provide the execution method.


The Skill Store Is Becoming the Distribution Layer of the Agent Economy

LibTV is not alone in recognizing the opportunity.

Throughout 2026, major technology companies, foundation-model providers, local-service platforms and content ecosystems have begun competing to build Skill Stores and Agent marketplaces.

The strategic logic is straightforward:

Whoever controls capability distribution can influence where users, creators and model traffic gather.

A successful Skill Store can determine:

  • Which Skills receive visibility;

  • Which creators gain users;

  • Which models receive calls;

  • Which technical standards dominate;

  • How revenue is divided;

  • Whether users remain inside the platform ecosystem.

This is similar to the position Apple gained through the App Store.

Apple did not merely distribute software.

It controlled discovery, installation, payment, updates, reviews and the commercial relationship between developers and users.

Skill Stores are competing for a similar role in the Agent era.


Why Compete for Stores That Are Mostly Free?

There is an obvious question.

Most Skill Stores still offer their Skills free of charge. Only a limited number of platforms are testing direct paid distribution.

Why are large companies competing so aggressively for marketplaces that do not yet generate significant direct revenue?

Because the Skill itself may not be the first source of profit.

The downstream activity is.

A single Skill invocation can drive:

  • Token consumption;

  • Model API calls;

  • Cloud-compute usage;

  • SaaS transactions;

  • Advertising exposure;

  • Platform retention;

  • Payment activity;

  • Enterprise-service demand.

Different platforms can monetize this activity in different ways.

Cloud providers may earn from compute.

Consumer ecosystems may earn from advertising and transactions.

Local-service platforms may increase engagement and conversion.

Content platforms may attract creators and retain traffic.

The store captures the user.

The ecosystem captures the revenue.


RMB 10 Million Is a Bid for the Supply Side

LibTV’s RMB 10 million incentive plan is therefore more than a creator-marketing campaign.

It is a bid for the supply side of the Agent economy.

A Skill Store without high-quality Skills is an empty mall.

Professional creators need reasons to invest time in structuring their experience, maintaining Skills and supporting users.

The incentive plan provides several possible benefits:

  • Official platform traffic;

  • Direct cash rewards;

  • Brand collaboration opportunities;

  • Greater visibility;

  • Potential long-term revenue sharing.

The immediate goal is to attract strong creators.

The long-term goal is to complete a commercial loop:

Creator expertise
→ Skill packaging
→ User invocation
→ Token consumption
→ Revenue sharing

The platform that establishes this loop first may gain one of the most valuable entrances to the Agent era.


The Missing Infrastructure: Who Prices and Settles the Tokens?

The Skill Store solves one important problem:

How should capabilities be packaged and distributed?

It does not solve another equally important question:

How should the Tokens and model resources consumed by those capabilities be priced, routed and settled?

Every professional Skill creates a chain of model activity.

A video-production Skill may need to:

  • Interpret a creative brief;

  • Generate a script;

  • Design storyboards;

  • Create images;

  • Produce video clips;

  • Review continuity;

  • Revise failed shots;

  • Generate voice and music;

  • Complete final quality checks.

Each step consumes Tokens or other forms of AI compute.

As Skills become more sophisticated, the cost of running them becomes increasingly important.

A Skill may be creatively excellent yet commercially unusable if the underlying model cost is too high.

This is where Unexhub enters the picture.


Unexhub: Turning Token Supply Into a Tradable Market

Unexhub is building a compute trading hub centered on Token aggregation, routing and distribution.

Its goal is to connect:

  • Large model providers;

  • Specialized AI suppliers;

  • Individual developers;

  • Small and medium-sized model companies;

  • Enterprise users;

  • Agent application platforms.

The platform is bringing together model access from providers such as Gemini, ChatGPT, Claude, DeepSeek and GLM, while also allowing smaller suppliers to distribute specialized or cost-efficient AI capabilities.

Instead of forcing developers to integrate every provider separately, Unexhub aims to provide a unified supply network.

In simple terms:

LibTV distributes what the Agent knows how to do.
Unexhub distributes the intelligence and compute required to do it.


From Fixed API Prices to Market-Based Bidding

Most model APIs still operate through fixed price lists.

The provider announces a rate, and the customer either accepts it or chooses another service.

But compute markets are not static.

Demand rises during peak periods and falls during quieter hours. Some providers face capacity shortages, while others have unused resources.

Unexhub is developing an API bidding mechanism intended to match supply and demand more dynamically.

This creates potential value for both sides.

For developers and enterprise users

During off-peak periods or when competitive supply is available, developers may access lower-priced model resources.

Under applicable market and service conditions, selected calls may cost as little as approximately 30% of prevailing official rates.

That can create substantial long-term savings for high-volume scenarios such as:

  • AI video production;

  • Batch content generation;

  • Customer-service Agents;

  • Enterprise knowledge bases;

  • Coding workflows;

  • Data processing;

  • Large-scale Skill execution.

Users pay according to actual Token consumption rather than maintaining unnecessarily high fixed commitments.

For AI and compute suppliers

Suppliers gain an additional distribution channel and can expose available capacity to real demand.

During high-demand periods, dynamic pricing may allow suppliers to earn stronger returns.

Unexhub also supports faster usage-based settlement, helping suppliers recover revenue according to completed calls rather than waiting through long billing cycles.

For smaller providers, this reduces cash-flow pressure:

Supply valid model capacity, receive the corresponding usage revenue more quickly.

A one-click channel connection also reduces the engineering and business cost of reaching new customers.


Early Supplier Benefits: Real-Time Settlement and 0% Commission

To attract high-quality supply, Unexhub offers early-stage incentives under applicable participation programs.

Eligible early suppliers may receive commission rates as low as 0%.

The practical value is clear.

A smaller provider entering a traditional distribution channel may lose a meaningful share of revenue before it has built stable usage.

A zero-commission starting period allows the supplier to:

  • Test market demand;

  • Build usage volume;

  • Improve service stability;

  • Acquire customers;

  • Recover infrastructure investment;

  • Preserve more early revenue.

Combined with real-time or accelerated settlement, the model reduces two of the biggest pressures facing smaller AI suppliers:

  • High channel costs;

  • Long payment cycles.

The platform is not only competing for users.

It is competing for supply.


Millisecond Routing Connects Skills to the Right Resources

A Token marketplace becomes useful only when requests can be matched with the right model quickly.

Unexhub’s routing layer is designed to make decisions based on factors such as:

  • Required model capability;

  • Current price;

  • Response latency;

  • Provider availability;

  • Context-window requirements;

  • Task category;

  • Region;

  • Enterprise policy.

Millisecond-level routing is especially important for Agent applications.

A professional video Skill may call several different models during one production:

  • A reasoning model for creative planning;

  • A language model for scriptwriting;

  • An image model for storyboarding;

  • A video model for clip generation;

  • A low-cost model for initial quality checks;

  • A premium model for final review.

Sending every stage to the strongest and most expensive model would make the workflow difficult to monetize.

Sending every stage to the cheapest model would undermine quality.

The routing layer needs to select the lowest-cost resource capable of satisfying each step.

That is how Token aggregation becomes a true operating system for Agent economics rather than a simple model directory.


LibTV Trades Capabilities; Unexhub Trades Resources

When LibTV and Unexhub are viewed together, the structure of the emerging Agent economy becomes clearer.

LibTV manages the capability transaction.

Creators package directing experience into Skills. Users select them. Agents execute them.

Unexhub manages the resource transaction.

Model suppliers provide Token capacity. Developers and enterprises purchase it according to demand. Prices and routes can respond to market conditions.

One platform answers:

How should this video be created?

The other answers:

Which models and compute resources should power it, and at what cost?

One solves application-layer distribution.

The other supports infrastructure-layer circulation.

Ecosystem LayerCore Role
Skill CreatorPackages professional experience
Skill StoreDistributes and monetizes capability
Agent PlatformExecutes the workflow
Model NetworkSupplies intelligence
Token MarketplaceMatches demand with compute supply
Settlement SystemDistributes revenue among participants

The two systems reinforce each other.

More valuable Skills generate more Token demand.

Lower Token prices make sophisticated Skills affordable.

Faster settlement attracts more suppliers.

More suppliers improve availability, price competition and model diversity.

This is the foundation of an ecosystem flywheel.


From “Tools” to “Transactions”

The most important change is not that AI tools are becoming easier to use.

It is that AI capabilities are becoming items that can be packaged, purchased, consumed and monetized.

The early generative-AI market primarily sold access to tools:

  • A chat interface;

  • An image generator;

  • A coding assistant;

  • A video platform.

The Agent economy introduces a transactional layer.

Users are not merely subscribing to software.

They can purchase or invoke specific capabilities:

  • Produce a luxury advertisement;

  • Migrate a codebase;

  • Review a contract;

  • Analyze financial statements;

  • Build a customer-service workflow;

  • Generate a short-form drama.

Creators can sell methodology.

Platforms can distribute demand.

Model providers can sell inference.

Compute suppliers can monetize capacity.

The unit of commerce shifts from software access to completed capability.


The 2008 App Store Parallel

When Apple launched the App Store in 2008, it did more than create a place to download software.

It aligned four elements:

  • Developers;

  • Users;

  • Distribution;

  • Payment.

Developers gained access to a global market.

Users gained a trusted place to discover applications.

Apple provided infrastructure and collected commissions.

A three-sided flywheel began to spin.

The Agent ecosystem is beginning to repeat the pattern.

Skill Stores form the application layer

Creators package professional expertise and list it for users.

Agent platforms form the execution layer

Agents interpret requirements and run the Skills.

Token markets form the resource layer

Model and compute providers supply the intelligence required for execution.

Settlement systems form the commercial layer

Revenue is distributed among creators, platforms and suppliers.

The difference is that the App Store sold software.

The Agent Store sells capabilities—and the compute required to deliver them.


What Still Needs to Be Solved?

The “App Store moment” is close, but several pieces remain incomplete.


Quality Assurance

A Skill with a professional title does not guarantee professional output.

Platforms need:

  • Creator verification;

  • Standardized evaluations;

  • User ratings;

  • Version control;

  • Performance reporting;

  • Clear quality guarantees.


Security

A Skill is executable logic and methodology.

A malicious Skill could:

  • Steal data;

  • Manipulate the Agent;

  • Call unauthorized tools;

  • Trigger unnecessary Token spending;

  • Send information to external systems.

Skill Stores need permission controls, supply-chain review and runtime monitoring.


Intellectual Property

A Skill inspired by a recognizable director or visual style may raise questions about:

  • Authorization;

  • Attribution;

  • Imitation;

  • Ownership;

  • Commercial use.

Platforms need clear rules governing what creators are allowed to package and sell.


Pricing Transparency

Users need to know not only the Skill price, but also the expected compute cost.

A Skill that costs $5 to purchase but consumes $100 in Tokens is not truly a $5 product.

Marketplaces should disclose:

  • Estimated Token use;

  • Likely model routes;

  • Maximum cost settings;

  • Premium-model requirements;

  • Refund and failure policies.


Revenue Sharing

The ecosystem must decide how value is divided among:

  • Skill creators;

  • Agent platforms;

  • Model providers;

  • Compute suppliers;

  • Distribution channels.

Without a sustainable revenue-sharing mechanism, professional creators will not contribute their strongest methods.


Portability

A Skill built for one platform may not work on another.

Open standards will be needed to prevent creators and users from becoming permanently locked into one ecosystem.


Why Token Discounts Matter to Skill Economics

Lower Token prices are not merely a user benefit.

They can determine whether the Skill marketplace is commercially viable.

Consider an AI advertising workflow.

If the underlying model usage costs $100 and the customer pays $120, almost no margin remains for:

  • The Skill creator;

  • The Agent platform;

  • The model supplier;

  • Customer support;

  • Sales and distribution.

If competitive routing reduces model costs to $30 or $40, the same workflow can support:

  • A lower customer price;

  • Creator royalties;

  • Platform fees;

  • Supplier revenue;

  • Sustainable operating profit.

This is why Unexhub’s pricing advantage matters.

Under applicable conditions, access at approximately 30% of official market rates can create the economic room required for the entire Agent transaction chain.

The discount does not simply reduce a bill.

It can finance the revenue-sharing model.


The Closed Loop That Everyone Is Racing to Build

The winning Agent platform will likely be the one that completes this loop:

Professional experience
→ Skill creation
→ Store distribution
→ User invocation
→ Token consumption
→ Usage settlement
→ Creator and supplier revenue

LibTV’s RMB 10 million incentive supports the first part of the chain by attracting Skill creators.

Unexhub’s supplier incentives, API bidding and real-time settlement support the resource side.

One pays to expand the capability supply.

The other lowers the cost of expanding Token supply.

The two pieces are beginning to converge.


Unexhub’s Strategic Position

Unexhub does not necessarily need to compete directly with every Skill Store.

Its larger opportunity is to become the Token distribution and settlement layer beneath many Agent ecosystems.

Video platforms need models and compute.

Coding platforms need models and compute.

Customer-service platforms need models and compute.

Legal, financial and marketing Agents all need:

  • Model access;

  • Routing;

  • Metering;

  • Price discovery;

  • Billing;

  • Supplier settlement.

By aggregating large model providers together with smaller specialized suppliers, Unexhub can provide a broader and more competitive supply network.

Its proposed value includes:

  • Unified access to Gemini, ChatGPT, Claude, DeepSeek, GLM and other models;

  • Access to specialized models from individual and SME suppliers;

  • Market-based API bidding;

  • Millisecond routing;

  • Pay-per-Token billing;

  • Selected prices as low as approximately 30% of official rates;

  • One-click supplier access;

  • Real-time usage settlement;

  • Early supplier commission rates as low as 0%;

  • Enterprise integration and customization services.

For developers, this reduces model-integration complexity.

For suppliers, it creates a faster path to demand and revenue.

For enterprises, it provides more flexible model choice and cost management without maintaining separate relationships with every provider.


The Flywheel Has Started to Turn

The emerging Agent economy may follow a simple flywheel:

  1. Experts package professional methods into Skills;

  2. Skill Stores distribute those capabilities;

  3. Users invoke them to complete real work;

  4. Every invocation consumes Tokens and compute;

  5. Token markets match demand with available supply;

  6. Settlement systems allocate revenue;

  7. Better returns attract more creators and suppliers;

  8. Greater supply reduces prices and improves quality;

  9. Better Skills and lower costs attract more users.

At this point, Agents stop being impressive demonstrations and begin becoming an economy.

The mobile-app revolution did not happen simply because phones became more powerful.

It happened because capability, distribution, infrastructure and payment aligned.

The same alignment is now beginning in the Agent market.


The Bottom Line

LibTV is investing RMB 10 million to attract Skill creators and has placed more than 100 director-grade capabilities inside a professional Skill Hub.

Unexhub is building a Token aggregation, routing, trading and settlement network beneath the Agent economy.

One competes for the supply of capability.

The other competes for the supply of resources.

One packages directors’ experience so Agents know how to create.

The other connects models and compute so Agents have the resources to execute.

The decisive question is no longer whether professional experience can be packaged.

It can.

Nor is it whether Agents can invoke that experience.

They increasingly can.

The remaining challenge is to complete the commercial cycle:

Capability packaging
→ Skill distribution
→ Agent execution
→ Token consumption
→ Instant settlement
→ Creator and supplier revenue

The company that completes this loop first may control one of the most important entrances to the Agent era.

In 2008, the App Store sold software.

In 2026, the emerging Agent marketplace sells professional capability—and the compute required to bring that capability to life.

The Agent ecosystem’s App Store moment has not fully arrived.

But the storefront, the creators, the resource network and the payment rails are already taking shape.

It may be much closer than the industry expects.


中文版

千万激励+百位AI导演:Agent应用生态,正在从“工具”走向“交易”

2026年7月14日,LibTV完成了一次可能比普通产品升级更具深远意义的动作。

随着Agent功能正式上线,LibTV同步推出其专业视频Agent Skill Hub。首批上线超过100项导演级视频Skill,覆盖影视导演、品牌广告、电商营销、短剧制作、MV及自媒体创作等多个领域。

每一项Skill封装的都不只是视觉预设,还可能包括:

  • 完整的创作工作流;

  • Prompt与指令策略;

  • 多模型组合逻辑;

  • 风格控制方法;

  • 分镜设计规范;

  • 剪辑与节奏规则;

  • 经过真实项目验证的行业经验。

创作者既可以直接调用已有Skill,也可以将自己的专业经验结构化,封装成一项新的能力,供其他用户调用。

LibTV随后推出了更具进攻性的生态计划:

总额1000万元的“Skill超创激励计划”。

优质Skill创作者有机会获得官方流量、现金奖励、品牌共创机会,以及面向未来的长期收益分成。

这已不只是推出一个新功能。

它是在搭建一个市场。


从“使用工具”到“购买能力”

在上一阶段的AI视频生产中,大部分工作仍然落在用户身上。

要生成一支能够交付的视频,创作者通常需要:

  • 编写复杂Prompt;

  • 反复调整参数;

  • 从多款模型中进行选择;

  • 尝试不同视觉风格;

  • 重新生成失败镜头;

  • 人工修正节奏;

  • 重新拼接素材;

  • 不断试错,直至结果可用。

用户名义上是在使用AI工具,实际上仍然承担着大量导演工作。

LibTV Agent试图提供另一种交互方式:

选择一项Skill,让Agent直接调用这套Skill背后的完整方法论。

创作者不必描述每一个镜头,而是可以直接选择一种导演级能力。

选择“韦斯·安德森式”Skill,Agent可以理解对称构图、精确调度与克制色彩。

选择古典武侠Skill,Agent可以调用空间关系、宽镜头节奏与传统视觉运动。

选择皮克斯风格广告Skill,Agent则可以结合情感叙事、角色节奏与品牌化的最终揭示。

用户不再需要逐步教Agent如何导演。

Skill已经把方法装进了系统。


Skill不是模板

Skill与传统模板之间存在本质区别。

模板是静态的。

参数、结构与输出方式大多已经固定,用户只是把自己的内容放入既有框架。

Skill则是动态的。

它能够封装一整套专业判断链:

  • 什么时候应该切镜头?

  • 哪种情绪适合暖色或冷色?

  • 产品应该多早出现?

  • 故事在什么位置需要放慢?

  • 当参考素材相互冲突时如何调整?

  • 脚本、分镜与最终审查应分别使用什么模型?

  • 尾帧应如何强化品牌记忆?

Agent不是简单复制一种视觉外壳,而是根据当前项目的具体情况,调用完整的创作方法。

这改变了AI视频生产的交互逻辑。

过去,用户需要不断和模型“说话”:

请生成这种氛围、这种灯光、这种构图、这种节奏和这种剪辑风格的视频。

指令越长,内容越容易相互冲突,跨镜头的一致性也越难维持。

有了Skill之后,用户可以在更高层表达:

按照这位导演的方法来完成。

Agent负责执行背后的完整决策链。


专业经验开始成为可交易资产

Skill Store真正的价值,并不只是提供更多风格。

它将专业经验转化为可以重复使用、传播和交易的数字资产。

传统创意行业中,一位导演最有价值的能力,通常藏在个人经验里:

  • 如何在前三秒建立氛围;

  • 什么时候展示产品;

  • 怎样平衡叙事与转化;

  • 如何控制节奏而不让观众流失;

  • 哪些细节能够制造高级感;

  • 怎样用Logo尾帧加强品牌记忆。

这些判断可能需要数年时间与几十个真实项目才能形成,却只能服务导演亲自参与的有限项目。

当方法论被结构化并封装成Skill之后,它就可以被全球用户反复调用。

这条转化路径可以概括为:

个人经验
→ 结构化方法论
→ Agent Skill
→ 商店分发
→ 持续收益

导演不再只能出售个人工作时间。

还可以分发专业判断的数字化版本。

因此,Skill不是普通工具组件,而是一项可以交易的能力。


Prompt工程不再是唯一入口

Skill Store的兴起,也意味着人们使用AI的方式正在变化。

生成式AI的第一阶段主要围绕Prompt工程展开。

用户尝试找到最准确的表达方式,编写更长的指令,并积累复杂的Prompt模板。

进入Agent时代,问题正在发生变化。

人们不再只问:

我应该如何向模型描述任务?

而会越来越多地问:

我应该安装或调用哪一项能力?

这是一种更具规模化潜力的交互方式。

多数用户并不想学习完整的导演术语、复杂的剪辑逻辑,以及几十款模型之间的技术差异。

他们希望选择一项已经得到验证的能力,然后直接获得成果。

同样的模式还可以延伸至更多领域:

  • 合同审查Skill;

  • 财务分析Skill;

  • 代码迁移Skill;

  • 智能客服Skill;

  • 网络安全响应Skill;

  • 医疗文档Skill;

  • 销售跟进Skill;

  • 跨境营销Skill。

Prompt负责表达目标。

Skill负责提供执行方法。


Skill Store正在成为Agent经济的分发层

LibTV并不是唯一看到这一趋势的平台。

2026年以来,大型互联网公司、基础模型厂商、本地生活平台与内容生态,都在加速建设Skill Store与Agent市场。

背后的战略逻辑非常清楚:

谁掌握能力分发入口,谁就能影响用户、创作者与模型流量最终聚集在哪里。

成功的Skill Store可以决定:

  • 哪些Skill获得曝光;

  • 哪些创作者获得用户;

  • 调用流量进入哪些模型;

  • 技术标准由谁制定;

  • 收益如何分配;

  • 用户是否长期留在平台生态。

这与苹果通过App Store获得的战略地位十分相似。

苹果不只是分发软件。

它还控制了发现、安装、支付、更新、评价,以及开发者与用户之间的商业关系。

Skill Store正在争夺Agent时代类似的位置。


Skill暂时免费,平台为何仍要争抢?

一个明显的问题是:

目前大部分Skill仍然免费,只有少数平台开始尝试付费Skill。

既然暂时赚不到多少直接收入,为什么大型企业仍在积极争夺这些“商店”?

因为Skill本身未必是第一层收入来源。

真正的价值可能来自下游消费:

  • Token消耗;

  • 模型API调用;

  • 云算力使用;

  • SaaS交易;

  • 广告曝光;

  • 用户留存;

  • 支付活动;

  • 企业服务需求。

不同平台可以通过不同方式完成商业化。

云厂商可以赚取算力收入。

消费生态可以获得广告与交易收益。

本地生活平台可以提高用户活跃与转化。

内容平台则可以吸引创作者并沉淀流量。

Skill Store负责获得用户。

生态负责实现收入。


1000万元,是对Skill供给侧的争夺

因此,LibTV推出的1000万元激励计划,不只是一次创作者营销活动。

它是在争夺Agent经济的供给侧。

没有优质Skill的Skill Store,就像一座没有商品的商场。

专业创作者需要足够的动力,投入时间整理经验、维护Skill并服务用户。

这套激励计划提供的可能权益包括:

  • 官方流量扶持;

  • 现金奖励;

  • 品牌合作机会;

  • 更高的平台曝光;

  • 未来长期收益分成。

短期目标,是吸引优质创作者。

长期目标,是跑通这条商业闭环:

创作者经验
→ Skill封装
→ 用户调用
→ Token消耗
→ 收益分配

谁能率先建立这套循环,谁就可能拿到Agent时代最有价值的入口之一。


缺失的基础设施:Token由谁定价、路由与结算?

Skill Store解决了一个重要问题:

专业能力应该如何被封装与分发?

但它没有解决另一个同样关键的问题:

这些能力被调用时消耗的Token与模型资源,应该如何定价、路由和结算?

每一项专业Skill都会带来一条模型调用链。

例如,一项视频制作Skill可能需要:

  • 理解创意需求;

  • 生成脚本;

  • 设计分镜;

  • 生成图片;

  • 制作视频片段;

  • 检查连贯性;

  • 修改失败镜头;

  • 生成语音与音乐;

  • 完成最终质检。

每一步都在消耗Token或其他形式的AI算力。

Skill越复杂,运行成本就越重要。

一项Skill即使创意效果出色,如果底层模型费用过高,也难以实现规模化商业应用。

这正是Unexhub进入市场的切入点。


Unexhub:让Token供给进入可交易市场

Unexhub正在搭建以Token聚合、调度与分发为核心的算力交易中心。

它希望连接:

  • 大型模型供应商;

  • 专业AI服务商;

  • 个人开发者;

  • 中小模型企业;

  • 企业用户;

  • Agent应用平台。

平台正在聚合Gemini、ChatGPT、Claude、DeepSeek、GLM等海内外主流模型,同时也允许中小型供应商提供专业化或更具成本优势的AI能力。

开发者不必逐家接入模型厂商,便可以通过统一供给网络调用不同资源。

简单来说:

LibTV分发Agent会做什么。
Unexhub分发Agent完成任务需要使用的智能与算力。


从固定API价格走向市场竞价

目前,大多数模型API仍然按照固定价目表收费。

供应商公布费率,用户选择接受或离开。

但算力市场并非静态。

高峰期需求上涨,低谷期资源可能闲置。一些供应商容量紧张,另一些供应商却有大量未被使用的算力。

Unexhub正在推进API竞价机制,希望按照市场供需更加灵活地匹配资源。

这能够同时为供需双方创造价值。

对开发者与企业用户

在闲时或竞争性供给充足时,用户可能以更低价格获得模型资源。

在符合市场与服务条件的情况下,部分调用价格最低可至市场官方价格的约 3折。

这一权益对于以下高频场景尤为重要:

  • AI视频生产;

  • 批量内容生成;

  • 智能客服Agent;

  • 企业知识库;

  • 代码工作流;

  • 数据处理;

  • 大规模Skill调用。

开发者可以按照实际Token消耗付费,而不必承担不必要的高额固定承诺。

对AI与算力供应商

供应商可以获得新的分发渠道,将可用容量直接匹配至真实需求。

在需求旺盛时,动态定价也有机会带来更高收益。

Unexhub还支持更快的按调用结算,帮助供应商根据实际完成的请求回收收入,而不是承受漫长账期。

对中小供应商来说,价值十分直接:

提供有效模型容量,就能更快回收对应的调用收益。

一键接入渠道也降低了供应商寻找新客户的技术与商务成本。


早期供应商权益:实时结算与最低0抽佣

为吸引优质供给,Unexhub还为符合条件的早期供应商设置了相应激励。

在适用计划下,早期供应商的渠道佣金最低可至 0%。

这一权益可以直接改善供应商的早期经营效率。

传统分发渠道往往在供应商尚未建立稳定调用量之前,就收取较高比例费用。

0抽佣启动期可以帮助供应商:

  • 验证市场需求;

  • 积累调用规模;

  • 优化服务稳定性;

  • 获取企业客户;

  • 回收基础设施投入;

  • 保留更多早期收入。

结合实时或加速结算,它可以缓解中小AI供应商面临的两类主要压力:

  • 渠道成本高;

  • 资金回款周期长。

平台争夺的不只是用户。

同样也在争夺供给。


毫秒级路由,把Skill与合适资源连接起来

Token市场只有在能够快速匹配资源时,才具有实际价值。

Unexhub的路由层可以根据以下因素作出选择:

  • 所需模型能力;

  • 当前调用价格;

  • 响应延迟;

  • 供应商可用性;

  • 上下文窗口;

  • 任务类型;

  • 所在区域;

  • 企业自定义策略。

毫秒级路由对Agent应用尤其重要。

一项专业视频Skill可能在同一项目中调用多款模型:

  • 推理模型负责创意规划;

  • 语言模型负责脚本;

  • 图像模型负责分镜;

  • 视频模型负责画面生成;

  • 低成本模型负责初步质检;

  • 高能力模型负责最终审查。

如果所有环节都使用最强、最昂贵的模型,工作流将难以商业化。

如果每一步都选择最低价模型,质量又会下降。

路由层需要为每个步骤选择能够满足要求的最低成本资源。

这正是Token聚合从“模型目录”走向“Agent经济操作系统”的关键。


LibTV交易能力,Unexhub交易资源

将LibTV与Unexhub放在一起,Agent经济的结构会更加清晰。

LibTV负责能力交易。

创作者把导演经验封装成Skill,用户选择,Agent执行。

Unexhub负责资源交易。

模型供应商提供Token容量,开发者和企业按照需求购买,价格与路由根据市场条件变化。

一个回答:

这支视频应该怎么拍?

另一个回答:

应该调用哪些模型和算力,以什么成本完成?

一个解决应用层分发。

另一个支撑基础设施层流通。

生态层级核心角色
Skill创作者封装专业经验
Skill Store分发并商业化能力
Agent平台执行工作流
模型网络提供智能
Token交易市场匹配需求与算力供给
结算系统向生态参与者分配收益

两套系统可以互相增强。

更优质的Skill带来更多Token需求。

更低的Token价格让复杂Skill更加可负担。

更快的结算吸引更多供应商。

更多供应商则会改善可用性、价格竞争与模型多样性。

这正是生态飞轮的基础。


从“工具”走向“交易”

最重要的变化,并不只是AI工具变得更容易使用。

而是AI能力正在被包装、购买、消耗并实现商业化。

早期生成式AI市场主要出售工具访问权:

  • 一个聊天界面;

  • 一款图像生成器;

  • 一个代码助手;

  • 一个视频平台。

Agent经济则增加了交易层。

用户不再只是购买软件订阅。

还可以直接购买或调用某项能力:

  • 制作一支高奢广告;

  • 迁移一个代码库;

  • 审查一份合同;

  • 分析财务报表;

  • 部署客服工作流;

  • 生成一部短剧。

创作者出售方法论。

平台分发需求。

模型供应商出售推理能力。

算力供应商将容量变现。

商业交易的最小单位,正在从软件使用权转向可交付能力。


2008年App Store的启示

2008年,苹果推出App Store时,并不只是创建了一个软件下载页面。

它将四项要素连接起来:

  • 开发者;

  • 用户;

  • 分发;

  • 支付。

开发者获得全球市场。

用户获得可信的软件发现渠道。

苹果提供基础设施并收取佣金。

三方飞轮开始运转。

Agent生态正在复刻类似结构。

Skill Store构成应用层

创作者封装专业经验,并将其上架给用户。

Agent平台构成执行层

Agent理解需求并运行Skill。

Token市场构成资源层

模型与算力供应商提供执行所需的智能。

结算系统构成商业层

收益在创作者、平台与供应商之间分配。

区别在于,App Store出售软件。

Agent Store出售的是能力,以及让能力得以运行的算力。


还需要解决哪些问题?

Agent生态的“App Store时刻”已经很近,但仍有多项基础问题尚未完成。


质量保障

拥有专业名称的Skill,不代表一定能稳定输出专业结果。

平台需要建立:

  • 创作者认证;

  • 标准化评测;

  • 用户评分;

  • 版本管理;

  • 效果报告;

  • 清晰的质量保障。


安全问题

Skill是一套可执行逻辑与方法论。

恶意Skill可能:

  • 窃取数据;

  • 操控Agent;

  • 调用未授权工具;

  • 制造不必要的Token消耗;

  • 将信息发送至外部系统。

Skill Store需要建立权限控制、供应链审查与运行时监控。


知识产权

以特定导演或视觉风格为灵感的Skill,可能涉及:

  • 授权;

  • 署名;

  • 模仿;

  • 所有权;

  • 商业使用。

平台需要明确创作者能够封装和出售哪些方法。


价格透明度

用户不仅要知道Skill价格,还需要知道预计的算力费用。

一项购买价5美元、运行时消耗100美元Token的Skill,并不是真正的5美元产品。

平台应该披露:

  • 预计Token消耗;

  • 可能调用的模型;

  • 最大费用设置;

  • 是否需要高价模型;

  • 失败与退款规则。


收益分配

生态需要明确以下角色如何分享价值:

  • Skill创作者;

  • Agent平台;

  • 模型供应商;

  • 算力供应商;

  • 分发渠道。

如果分配机制无法持续,专业创作者不会贡献最优质的方法论。


可迁移性

在某个平台创建的Skill,未必能在另一平台运行。

行业需要开放标准,避免创作者与用户被永久锁定在单一生态中。


为什么Token优惠决定Skill经济能否成立?

Token成本下降不仅是用户权益。

它可能直接决定Skill市场能否形成可持续商业模式。

假设一套AI广告工作流的模型成本是100美元,而客户只愿意支付120美元,那么剩余空间几乎无法覆盖:

  • Skill创作者分成;

  • Agent平台服务费;

  • 模型供应商收益;

  • 客户支持;

  • 销售与分发。

如果通过竞争性供给与高效路由,把模型成本降至30或40美元,同一套工作流便可以同时容纳:

  • 更低的客户价格;

  • 创作者版权收益;

  • 平台费用;

  • 供应商收入;

  • 可持续利润。

因此,Unexhub的价格优势不只是让账单更低。

在适用条件下,约为官方市场价格3折的成本空间,可以为整个Agent交易链提供收益分配所需的利润基础。

优惠本身,也可能成为商业闭环的一部分。


所有人都在争夺的闭环

最终胜出的Agent平台,很可能是率先跑通以下链路的平台:

专业经验
→ Skill创建
→ 商店分发
→ 用户调用
→ Token消耗
→ 用量结算
→ 创作者与供应商收益

LibTV的1000万元激励,正在吸引Skill创作者,补充链路的前半部分。

Unexhub的供应商激励、API竞价和实时结算,则支撑资源供给侧。

一个用资金扩大能力供给。

另一个降低Token供给的进入门槛。

两块拼图正在靠拢。


Unexhub的战略位置

Unexhub不一定需要直接与每一家Skill Store竞争。

它更大的机会,是成为多个Agent生态背后的Token分发与结算层。

视频平台需要模型与算力。

编程平台需要模型与算力。

智能客服平台同样需要模型与算力。

法律、财务、营销等Agent应用都离不开:

  • 模型接入;

  • 任务路由;

  • 用量计量;

  • 价格发现;

  • 账单管理;

  • 供应商结算。

通过聚合大型模型厂商与中小专业供应商,Unexhub可以构建更加广泛、竞争更加充分的供给网络。

其价值包括:

  • 统一接入Gemini、ChatGPT、Claude、DeepSeek、GLM等主流模型;

  • 接入个人与中小供应商提供的专业模型;

  • 市场化API竞价;

  • 毫秒级路由;

  • 按Token实付;

  • 部分调用最低可至官方价格约3折;

  • 供应商一键接入;

  • 实时调用结算;

  • 早期供应商最低0抽佣;

  • 企业级接入与定制服务。

对开发者而言,这能降低模型对接复杂度。

对供应商而言,这能缩短连接需求与实现收益的路径。

对企业而言,则可以在不分别维护每家供应商关系的情况下,获得更灵活的模型选择与成本控制能力。


飞轮已经开始转动

Agent经济未来可能形成这样一套飞轮:

  1. 专业人士将方法封装为Skill;

  2. Skill Store完成能力分发;

  3. 用户调用Skill完成真实工作;

  4. 每次调用消耗Token与算力;

  5. Token市场匹配需求与可用供给;

  6. 结算系统完成收益分配;

  7. 更好的收益吸引更多创作者与供应商;

  8. 更充分的供给降低价格并提高质量;

  9. 更多Skill与更低成本继续吸引用户。

到了这一阶段,Agent就不再只是令人惊叹的演示。

它会开始成为一种经济形态。

移动应用革命并不是因为手机仅仅变得更强。

而是因为能力、分发、基础设施与支付在同一时间完成了连接。

Agent市场正在出现同样的对齐。


结语

LibTV投入1000万元吸引Skill创作者,并将超过100项导演级能力放进专业Skill Hub。

Unexhub则正在搭建Agent经济底层的Token聚合、路由、交易与结算网络。

一个争夺的是能力供给。

另一个争夺的是资源供给。

一个把导演经验封装起来,让Agent知道应该如何创作。

另一个连接模型与算力,让Agent获得真正执行所需的资源。

如今,决定性问题已经不再是专业经验能否被封装。

答案是可以。

也不再是Agent能否调用这些经验。

它们已经越来越擅长。

最后的挑战,是完成完整商业闭环:

能力封装
→ Skill分发
→ Agent执行
→ Token消耗
→ 即时结算
→ 创作者与供应商收益

谁能率先跑通这条链路,谁就可能掌握Agent时代最重要的入口之一。

2008年,App Store出售的是软件。

2026年,正在形成的Agent市场出售的是专业能力,以及驱动这些能力运行的算力。

Agent生态的“App Store时刻”还没有完全到来。

但商店、创作者、资源网络与支付管道,都已经开始成形。

它可能比整个行业预想得更近。