Cetus Skills give AI agents a documented way to interact with Cetus’s trading and liquidity infrastructure on Sui, bringing the project’s AI ambitions into a public developer toolkit.
The repository contains Claude Code skills for three parts of the protocol: its swap aggregator, concentrated liquidity market maker, or CLMM, and dynamic liquidity market maker, or DLMM.
For developers, the appeal is straightforward. An agent needs detailed instructions about how a protocol works before it can reliably prepare an action. Cetus packages those instructions around existing interfaces, helping developers connect AI workflows with on-chain operations.
The important distinction is between preparing a transaction and authorizing it. Access to trading tools does not, by itself, give an AI agent permission to move a user’s funds.
What Cetus Skills Actually Enable
The toolkit’s components serve different purposes. One focuses on finding swap routes, while the other two address liquidity provision.
Component | Documented capabilities | Practical purpose |
|---|---|---|
Retrieve quotes, compare routes and build unsigned swap transactions | Prepare token exchanges | |
Create pools, open and close positions, adjust liquidity and collect fees | Manage liquidity within selected price ranges | |
Manage liquidity across price bins, use distribution strategies and collect fees | Configure liquidity allocation with dynamic fees |
These are developer capabilities rather than evidence that an autonomous trading service is already operating at scale.
The public repository also includes an SDK dependency snapshot checked on October 21, 2026. That provides a dated reference for the integration examples, although it should not be treated as proof of the toolkit’s original launch date.
A Swap Quote Still Needs a Signature
The aggregator documentation offers the clearest example of how the workflow operates.
An application collects the token pair, amount, wallet address and acceptable slippage. It then requests a route and presents the resulting quote. After confirmation, the service builds an unsigned transaction.
The wallet must subsequently sign that transaction before it can be submitted to Sui.
That separation matters. A system can use AI to research a swap and prepare the transaction while keeping final approval with the user. Developers could also build more automated arrangements, but those would need their own authorization rules.
The documentation identifies another limitation: transaction building currently supports fixed-input swaps. Fixed-output requests can generate quotes, but those quotes cannot be used through the same transaction-building flow.
Bitnxt’s guide to how AI agent wallets work explores the wider permission question: what an agent can suggest and what it can actually authorize are separate decisions.
Liquidity Management Brings More Decisions Into the Workflow
Swaps are only part of the toolkit’s scope.
The CLMM skill covers concentrated liquidity positions with custom price ranges. It also documents pool creation, liquidity adjustments and fee collection.
For an agent, this creates a more demanding task than choosing a token pair. A liquidity position requires decisions about where capital should be deployed and when it should be moved.
A hypothetical agent might monitor a position and prepare an adjustment when market conditions change. That illustrates a possible application; it is not a verified Cetus trading strategy.
Automating those steps can reduce manual work. It cannot establish that the selected range will produce a better financial outcome.
The DLMM skill takes a different approach. Its documentation describes liquidity distributed across discrete price bins, with Spot, Curve and BidAsk strategies. Supported operations include adding or removing liquidity, querying positions and collecting fees or reward tokens.
Those options give builders several ways to structure an application. They also make accurate transaction previews important: the user needs to understand where the funds will go, rather than simply accept an agent’s summary.
Where the AI Integration Adds Value
DeFi protocols already expose software interfaces. The potential contribution of skills is to make those interfaces easier for an AI coding agent to understand and use consistently.
Consider a developer building a portfolio assistant. The application could start with read-only tasks, such as retrieving market information or examining a position. It could then prepare an action for review.
This creates a useful progression from information to execution without requiring every workflow to begin with full autonomy.
The broader idea also connects with agent-to-agent crypto payments, where software needs clear instructions and permissions to transact. Cetus’s documented toolkit specifically addresses its DeFi interfaces; it does not establish that every broader agent-payment use case is supported.
What the Announcement Does Not Establish
The materials reviewed do not provide verified figures for active agents, agent-generated trading volume or investment returns attributable to Cetus Skills.
They also do not establish an Anthropic commercial partnership. Compatibility with Claude Code is a technical description, and should be reported that way.
There is similarly no basis in these materials to claim that the toolkit automatically creates additional demand for the CETUS token. That would require evidence connecting usage with token economics.
For Sui, the integration adds another developer access point. It sits alongside separate efforts to improve payment usability, including the developments discussed in Bitnxt’s coverage of zero-gas-fee stablecoin transfers. Those payment arrangements should not be assumed to apply to Cetus transactions.
Bitnxt View: Watch Execution Quality After the Announcement
Cetus has provided something concrete for developers to inspect: documented interfaces covering quotes, transaction preparation and liquidity operations.
The next meaningful evidence would show how applications use them. Relevant measures include successful execution rates, failed transactions, transaction costs and how clearly users can review proposed actions.
Agent activity also needs careful interpretation. More automated calls do not necessarily mean more users, more useful trading or better outcomes.
You might also like: Could AI agents become crypto’s biggest users?
Bitnxt’s assessment is that the toolkit deserves coverage as developer infrastructure. Its significance will become clearer when working integrations demonstrate reliable execution and understandable permissions—not simply when more projects describe themselves as AI-powered.
