The article mainly introduces claude-for-financial-services, a financial skills package officially released and open-sourced by Anthropic.
Article author and source: Frontiers in AI Technology
Investment bank analysts have a classic nightmare.
At 5 PM on Friday, MD sent me a message: "I need a pitch deck on Monday morning. Pull up the comps and run the DCF."
And then your weekend is gone.
Recently, Anthropic officially launched a repository called claude-for-financial-services, which compiles skill packages for the four most expensive sectors on Wall Street—investment banking, stock research, private equity, and wealth management—into Claude packages that can be directly installed.
Repository address: github.com/anthropics/financial-services
It's licensed under the Apache 2.0 license, entirely in Markdown and YAML, with no build steps; you can modify it directly after forking it.
Let's first explain what it is, and then explain why it's worth paying attention to.
The entire repository is divided into two layers: 11 end-to-end agents and 7 underlying skill packages for vertical industries.
Each agent in the layer is self-contained; once installed, it can run an entire pipeline without needing to install a bunch of dependencies first.
Let me mention a few of the most interesting ones separately.
The Pitch Agent takes comparable companies, precedent transactions, and LBO assumptions as input and outputs a pitch deck in a branded format. Previously, if the MD asked you to submit your pitch on Friday, you would have to start collecting components, running DCF, and building the deck by Wednesday. This process has now been streamlined into a single command.
Model Builder allows you to run DCF, LBO, and three-table models directly in Excel. It doesn't just export a table; Claude actually writes formulas and modifies cells within your .xlsx file.
GL Reconciler helps you find general ledger breaks, trace the root cause, and go through the signature process.
Earnings Reviewer : Earnings call and announcement features, automatic model updates, and research report drafting.
The official boundaries are clearly defined: these agents draft working papers for analysts, do not make investment decisions, do not execute trades, and every output is simply left there for human signature.
This positioning is very clever. The most sensitive thing in the financial industry is responsibility. Being able to generate revenue without taking the blame is the most realistic approach to B2B implementation.
The underlying Skill package is the part that's truly worth collecting.
The core package financial-analysis includes all common modeling skills, such as /comps comparable company analysis, /dcf valuation plus WACC plus sensitivity analysis, /lbo leveraged buyout model, and /3-statement-model three-statement model filling.
There's also audit-xls, which turns the analyst's biggest pain point, "the boss throws at you an Excel file that hasn't been maintained for ten years and asks you to find out which number is wrong," into a skill. Formula tracing, hard-coded detection, and balance verification can all be done with one click.
I was stunned when I saw this because the scene was so realistic.
The investment banking vertical package includes /cim information memos, /buyer-list potential buyer lists, and /merger-model share issuance dilution analysis. The equity research package includes /earnings earnings reviews, /initiate initial coverage reports, and /morning-note morning meeting minutes. The private equity package includes /ic-memo investment committee memos and /value-creation 100-day post-investment plan plus EBITDA bridge.
There are also 11 MCP data connectors, including Daloopa, FactSet, PitchBook, S&P Global, Morningstar, and Refinitiv.
Please note that the README clearly states that MCP access may require a data provider subscription or API key. Anthropic only provides the interface; you still need to pay for the data yourself. These are institutional seats, which individuals cannot afford. For domestic scenarios, you also need to connect to domestic data sources yourself.
Installation method: Claude Code three-line command:
Once installed, the forward slash command is available immediately; simply type /comps , /dcf , /earnings , and /ic-memo .
Let me talk about what I think is the greater significance of this matter.
Anthropic isn't promoting a product this time; it's setting a standard.
This elevated the Skill system from a "small tool tinkered with by community developers" to a " production-grade reference implementation for financial institutions ."
For those working in finance, the skill names are so easy to match with the tasks at their workstations that they can be used directly or modified from existing ones.
For those who want to learn how to write a "usable" Claude Skill, the official documentation is the best model. Go through the directory structure of plugins/agent-plugins/pitch-agent/ and see how Anthropic organizes system prompt, skill, and subagent. It's more useful than reading ten blog posts.
Repository address: github.com/anthropics/financial-services
