Tencent, Alibaba and ByteDance are locked in a fierce battle over the Skill Store

Skill is becoming one of the hottest keywords in the AI field.

Skill can be understood as an "operation manual" for AI Agents. It is a structured instruction file that clearly states which tools to call, how to make judgments in different situations, and what criteria to use for the final output. Once an Agent reads this file, it can execute tasks according to the preset path.

For example, a senior product manager can encapsulate the entire process of writing product requirement documents into a Skill. Once installed on anyone's Agent, it can output a standard requirement document following the same framework.

As the number of Skills increases, distribution platforms have emerged. Initially, developer communities such as GitHub and ClawHub took on this role. The uploading, searching, and downloading of Skills were all completed within these technical communities.

Large companies are also quickly following suit. In March this year, Tencent, Alibaba, and ByteDance successively launched Skill stores on their own Agent platforms. In the following two months, Zhipu, Meituan, and Xiaohongshu joined the fray. Internet giants, large model companies, local life service giants, and even content platforms are all vying for this entry point.

The essence of the competition for Skill stores is to secure a position as a traffic entry point in the AI era. Whoever controls the distribution rights controls the users.

However, except for ByteDance's Kouzis which has experimented with paid Skills, all other platforms offer free versions. Why are companies still competing for these "stores" that don't seem to make money?

01 Three Types of Players with Different Motives

Who is getting involved? Why is the Skill store worth competing for?

Before answering this question, let's look at a proven model.

In the era of mobile Internet, Apple's App Store doesn't just make money from the 30% download commission. Its core value lies in the fact that developers create apps to enter the iOS ecosystem, and users stay in the iOS ecosystem to use these apps, and then continue to consume within the ecosystem: buying iCloud, subscribing to Apple Music, and making in - app purchases. The distribution right is the entry point, and ecosystem consumption is the source of revenue.

The competition for Skill stores follows the same logic. Users will stay in the corresponding ecosystem to consume services based on where they are accustomed to obtaining Skills. The difference is that this logic has been verified in the mobile Internet era, while the Skill store is still in the "promising" stage. Understanding this, let's look at the different strategies of the three types of players.

The first type is Internet giants, which use Skill stores to attract traffic and make money within the ecosystem.

Alibaba has built the "Xiaoxiaobao" Skill market into its JVS Claw Agent. Users can synchronize the selected Skills to the tool with one click. The Skill market itself is free, but users need to consume computing power when using Skills, and computing power is part of Alibaba's cloud service revenue.

ByteDance is taking two parallel paths. Volcengine has launched Find Skill, targeting enterprise customers and integrating Skills from multiple sources such as ClawHub and GitHub. Kouzis' Skill store is for ordinary developers, lowering the threshold for creation and use and also supporting Skill sales. The goal is to attract developers and drive cloud service and computing power consumption through Skills.

Tencent's strategy is slightly different. SkillHub is essentially a localized mirror site of the overseas ClawHub, serving the functions of attracting traffic and local adaptation. However, Tencent's real trump card is the WeChat Mini - Program ecosystem. Relying on the mature service links accumulated by millions of mini - programs, Tencent can encapsulate various offline and online services into standardized Skills. If this path is successful, the business model will be similar to that of mini - programs, earning transaction commissions and advertising revenue.

Meituan uses the Skill ecosystem to support its main business. In April, it launched xia345, positioned as an AI Agent ecosystem navigator, which has included more than 20 Agents and over 7,000 Skills. In May, it conducted a public beta of an AI community called Miyou, with over 3,000 Agents and more than 40,000 Skills. From the navigator to the community, users see the sharing on "Miyou" and then go to "xia345" to download and use. Although Skills themselves don't make money, they can increase users' stay time in the Meituan system and create more conversion opportunities for core businesses such as in - store services and food delivery.

The second type is large model companies, which use Skill stores to retain users and make money from model calls.

In April, Zhipu launched the AgentMore Skills Square on its Agent platform Auto Claw, integrating three modules: official selection, Skill Hub, and open - source communities, and supporting one - click zero - Token installation.

Yuezhianmian took action earlier. In February, it launched Kimi Claw. Users can deploy Open Claw on the web - end with one click and are provided with a skill library. Users can directly install and call various Skills in the browser.

It seems most logical for large model companies to engage in Skill distribution. The model itself is the foundation for Skill operation. Developing a Skill store can drive continuous calls to their own large models and keep users within their own domain.

He Yu, an Agent engineer from a large model company, mentioned that self - developed Skills have a higher degree of compatibility with their own underlying models and offer a better user experience. Essentially, Skills are the "bait", and the model call volume is the "fish".

The third type is content platforms, which regard Skills as a new content category and make money from traffic and advertising.

Xiaohongshu recently launched Red Skill, which is currently in internal testing. Users can attach Skill links below posts, and clicking on them allows for copying and installing instructions. Different from the traditional Skill distribution process from search to configuration, Xiaohongshu takes a content recommendation approach, turning Skills into a content form that can be browsed and recommended. Xiaohongshu doesn't make money from Skills themselves but from the traffic and advertising revenue brought by this content.

The logic of the three types of players is the same: The Skill store itself doesn't make money, but it is an entry point to acquire and retain users. The real revenue lies beyond Skills.

However, this judgment is based on the premise that developers and users are really willing to use Skills.

Shansen Nan, an independent development blogger, mentioned that the Skill stores embedded in large - company products may not be as attractive as expected. They are more like ancillary functions within the overall product, with a weak presence and not being the main focus of large companies. Content platforms' natural dissemination ability gives them more competitiveness in the Skill distribution process.

In other words, although the stores are set up, they lack sufficient appeal.

02 Where are the Bottlenecks in the Skill Store Business?

To determine whether the Skill store business is profitable, the most direct way is to see if it makes money.

Currently, only ByteDance's Kouzis supports Skill transactions, where creators can set prices for their Skills. Other platforms almost all offer free distribution. What can really be called "transactions" are actually some people reselling open - source Skills in bundles on Xianyu by taking advantage of information asymmetry.

The so - called Skill "store" is just a metaphor for now. What's the problem?

The first hurdle is that it's difficult to price Skills.

The success of the App Store relies on a complete evaluation system: clear functions, stable experience, ratings, and user reviews. More importantly, the performance of the same App is consistent for everyone.

Skills lack this kind of certainty. The output of a Skill may vary greatly with different models and context environments. Shansen Nan told "AIX Finance" that different Agent products have different performance and the models they carry also vary. The results of running the same Skill on different products and models are uncontrollable. Even on the same product and model, due to the randomness of AI, the output may not be consistent.

He Yu added another perspective: Most general Skills for ordinary users have open - ended outputs without a single standard answer, and the industry currently lacks a unified standard for evaluating the effectiveness. High - quality Skills cannot be effectively identified, and users' screening costs are extremely high.

Without stable performance, an evaluation system cannot be established. Without an evaluation system, users lack a basis for paying.

The second hurdle is the opaque cost.

To complete the same task, the Token consumption of different Skills may vary by several times, but users have no way of knowing this before installation. For two Skills with the same function, which one is more "Token - saving"? There is no way to compare.

He Yu gave an example. He once used two long - text summarization Skills on the same platform. When processing the same document and giving the same instructions, the Token consumption differed greatly, and this difference was completely invisible when choosing the Skills. Users pay for Skills and also have to bear the uncertain Token consumption cost. How can this cost be calculated?

The third hurdle is the security risk.

Since this year, there have been cases of Skill poisoning. Malicious Skills are uploaded by imitating the names of popular Skills to steal user data. Although platforms have successively launched review mechanisms, this has also raised the threshold for developers to upload Skills.

Shansen Nan encountered restrictions when uploading Skills on Xiaohongshu. The platform only allows the upload of Markdown and TSD files, so complex Skills cannot be fully uploaded and can only be downgraded to a Prompt. A balance has not yet been found between security review and developer experience.

The last hurdle is the lack of a standardized protocol.

Different developers describe the same task in different ways, which easily leads to misunderstandings by the model and uneven execution results. He Yu said that the ambiguity in description makes it difficult to control the actual experience of Skills, and "usefulness" becomes unpredictable.

Coupled with the lack of standardized permission boundaries, the ideal effect of "one - time development, multi - platform distribution" cannot be achieved.

These four hurdles actually point to the same reason: Skills are essentially personalized workflows and naturally resist standardization. However, standardization is the prerequisite for commercialization.

Therefore, the current Skill stores are more like display shelves. The products are on display, but users don't know which one to choose, and they don't know if the chosen one will work well. There is still a long way to go before real "transactions" can take place.

03 How Far is the Skill Store from the App Store?

Let's shift our focus from platforms to developers.

Chen Xu, an independent developer, once uploaded a paid Skill on Kouzis. Six people paid on the day the Skill passed the review, and the homepage recommendation brought continuous exposure. However, the good times didn't last long. He soon found that he no longer had the opportunity to be recommended on the homepage. Users had to search actively to find his Skill, and he couldn't invest in traffic. The homepage exposure opportunities are completely controlled by the platform and are highly random.

This shows at least two things: First, there is a real demand for paid Skills; second, on existing platforms, developers' distribution capabilities are extremely limited.

So, can the Skill store become the next App Store? Currently, there are two main obstacles.

On the one hand, there is no unified evaluation system for Skills. Chen Xu mentioned that he usually chooses Skills based on the star count on GitHub because these Skills have been tested by real users. However, the popularity rankings on domestic platforms deviate from those on foreign platforms, and the indicators may be inaccurate. Without a cross - platform and standardized evaluation system, users can only choose by luck.

On the other hand, Skills have strong personalized attributes. Shansen Nan mentioned that the effectiveness of most general Skills on the market is limited. Truly useful Skills need to be tailored to individual workflows, repeatedly adjusted in actual work, and precipitate a unique methodology. For example, even two Skills both labeled as "writing assistants" may have completely different adapted workflows and output styles.

Without an evaluation system, the Skill store can only remain at the stage of a display shelf.

However, from another perspective, Skills are essentially a new form of commodity. In the past, users paid for "certainty". When they needed a function, they downloaded an App. Now, they are paying for "possibilities", for a creative ability and a reusable methodology.

He Yu divided the scenarios with a payment basis into two categories: one is office necessities, such as contract review and data report generation, where enterprises have a strong willingness to pay; the other is personal tools, such as resume optimization and writing of study - abroad documents, where the payment conversion rate is relatively high.

The question is, who can turn this potential into a real business?

The three types of players have their own advantages but also their own shortcomings.

Internet giants are closest to the application scenarios, but Skill stores are just an "add - on" for them, and they won't invest core resources. Large model companies have a natural advantage in model adaptation, but their ecosystems are not as strong as those of giants. Skill stores are just value - added services, and their essence is to encourage users to continuously call their models. Content platforms have the strongest dissemination ability. In the stage when Skills don't have a standardized evaluation system, users rely on blogger recommendations and usage demonstrations to choose Skills, which is exactly what content platforms are good at. However, they are the farthest from the technical ecosystem.

The instability, personalized attributes, and security risks of Skills make this business much more difficult than it seems. For the players involved, none of them has yet made "buying Skills" as natural as "buying Apps".

This article is from the WeChat official account "AIX Finance". Author: AIX Finance Team. Republished by 36Kr with permission.