Make your AI defi native: a true expert on onchain finance.

It can now read a wallet. Give it an address and it tells you what the wallet holds, how much of that sits in DeFi and what it owes against it, and whether the wallet is up or down, with every number dated. Holdings and profit and loss come from Zerion's API with your own free key, then each vault, loan, or pool position goes through the same look-through as any product. Read-only as always: a dollar-cost-averaging request ends at a preview that needs your approval, never at a trade.

Hardened for agent catalogs. The skill never loads newer files from the internet at runtime; what you reviewed is what runs, and updates only arrive through your install channel. Everything it fetches is treated as data, never as instructions. Paid API calls need your explicit yes before any money is spent, keys stay out of chats and logs, and connecting any data server requires your approval first.

It can now tell you whether a tokenized fund's yield is real. Putting money into a tokenized treasury or credit fund is not like buying a token on an exchange. Someone decides the price you come in at, and when they decide it can work for you or against you. This covers the four ways that settlement actually happens, how to tell whether an advertised yield was measured or simply announced, and what the issuer takes before any of it reaches you. Contributed by @mishablank.

It can now read market structure. How to tell a real short squeeze from a crowded trade. How to spot wash trading and other manipulation from traces left onchain. Why a tokenized stock has three different prices at the same time, and which one actually matters. What people mean when they call a token liquid, and how to check. Also added: how attention turns into an asset, a new kind of lending market where your interest rate is really a bid for your place in the exit queue, and how to read raw blockchain logs when the dashboards do not go back far enough.

Went through every data source by hand. Twelve new ones, thirty-one in total. Fixed the logins, the endpoints that had quietly moved, and one source that was promising more than it could actually answer.

Taught it where to look things up. Each kind of question now maps to the right data source, with the common mistakes attached, plus a note on what your own API key would unlock if you have one.

No hand-waving in comparisons. Every option it puts in front of you has to show its own numbers, not just the one it recommends. A test enforces it.

Options, trades, and how to judge a manager. Three new references: why providing liquidity is really selling an option and what that costs you, how discounted locked-token deals work with three real 2026 examples, and eleven professional curators' published frameworks boiled down to one checklist you can score any manager against.

Four different AI models tried to break it. Everything they found got fixed, including a clearinghouse process described in the wrong order and a price that was off by a factor of ten.

Caught in real use. One lookup was only returning half of Morpho's vaults, so the skill was silently missing entire sets of positions without saying so.

First public release. Tested by running the same questions with and without the skill and comparing, reviewed by independent AI models, and every source checked by hand.