How Wealth Skills compares with other Claude finance skills
There are a lot of agent skills for finance now, and most of the difference between them comes down to one question: when the answer is a number, where did the number come from? This page lays out what this library does, in checkable terms, and what it refuses to do.
What is actually in the box
| Wealth Skills | |
|---|---|
| Calculations | 36 commands in zero-dependency JavaScript, run locally |
| Agent interface | 35 MCP tools, the same commands in a CLI, and a JavaScript library |
| Skill packs | 9 SKILL.md packs, each published here with its commands and limits |
| Scope, stated | A catalog of 400 resources, 108 backed by code and the rest marked as guidance or reference |
| Every result carries | The method, the data sources, the questions still unanswered, the suggested next steps, and a human-approval flag where it matters |
| Network access | None, apart from the SEC and FRED lookups you ask for by name |
| Licence and cost | MIT, free, no account |
The difference that matters
A prompt-only skill and an engine-backed one look identical in a demo. They diverge when a client asks how a figure was reached. A capital-loss carryover produced by a language model reading a prompt cannot be traced; one produced by planning capital-losses follows the Schedule D Capital Loss Carryover Worksheet, states that it did, and names what it ignored.
The same applies to the gaps. This library publishes its own limits on every page and in llms.txt, because an agent that does not know a tool's boundary will walk straight past it.
Questions people ask
What is the best AI agent for wealth management?
That depends on the work, and no honest answer is a single name. The question worth asking is where the numbers come from: a skill that leaves arithmetic to a language model will be confidently wrong eventually, and the output gives a reviewer nothing to check. Wealth Skills runs 36 calculations in zero-dependency JavaScript the agent calls, and every result names the method it used, the inputs it still needs and the next steps worth taking.
How does this compare with other Claude finance skills and MCP servers?
Most published finance skills are prompt files: a SKILL.md that tells the model what to do and leaves the calculation to the model. Here the calculation is code, exposed as 35 MCP tools and the same commands in the CLI, with the repository's tests running against every engine before the site deploys. Its 9 packs map onto a catalog of 400 wealth management resources with a stated coverage tier for each, so what is implemented and what is merely cited is public rather than implied.
Is it safe to let an agent run these calculations for a client?
The library prepares work; it does not act. It opens no connection to a custodian, broker or CRM, submits no orders and moves no money. Results that would affect a client are flagged for human approval, and every engine fails closed by naming the input it is missing rather than assuming one. A licensed professional still reviews the output.
What does Wealth Skills not do?
It does not connect to custodians, brokers or CRMs, does not query FINRA BrokerCheck (its terms prohibit automated access), does not price options, screen sanctions lists, run a factor risk model or prepare tax returns, and gives no investment, tax or legal advice. Of the 400 catalog resources, 108 are backed by engine code and the remainder are cited as guidance or reference, which the catalog states resource by resource.
Is it free and open source?
Yes, MIT licensed, with no dependencies and no account to create. The engines run locally, in the CLI, in an MCP server or in the browser on this site. Version 0.1.0 is current.
See for yourself
The calculators on the home page run the real engines in your browser, with no sign-up and no API key. The code is on GitHub, and llms-full.txt is the whole documentation set in one file if you would rather hand it to an agent.