Best AI Agent Skills for Financial Modeling (2026)

If you build financial models for a living, you already know the tedious parts: pulling numbers out of PDFs by hand, rebuilding the same DCF structure for the fifth time this quarter, hunting for the comp data that justifies your assumptions. Claude Code skills target exactly those parts. They don't replace your judgment on the model. They cut the hours of grunt work that surround it.

A skill is a folder with a SKILL.md file that teaches Claude a specific job, plus any scripts it needs. When the task matches, Claude loads the skill and follows it. For finance work, four skills do most of the heavy lifting. Here's what each one is for, how to install it, and what to check before you trust it with your numbers.

xlsx-spreadsheet: models with real formulas, not pasted values

The xlsx skill is the core one for modeling. It builds Excel files where the cells hold actual formulas, so your DCF or three-statement model stays linked and auditable instead of being a grid of hardcoded numbers. You can ask it to build a revenue buildup, wire the income statement to the balance sheet, or add a sensitivity table, and it writes the formulas a reviewer can trace.

Install it by dropping the skill into your Claude Code skills directory:

git clone https://github.com/anthropics/skills ~/.claude/skills-temp
cp -r ~/.claude/skills-temp/document-skills/xlsx ~/.claude/skills/xlsx-spreadsheet

Or grab just the xlsx folder from the official Anthropic skills repo at github.com/anthropics/skills and place it under ~/.claude/skills/. Restart Claude Code and it picks up the skill automatically.

The thing to verify: outputs are easy to check, formula logic is not. Always open the model and confirm the links go where you intended, especially around circular references like interest on a revolver. The skill writes correct formulas most of the time, but a wrong cell reference in a financial model compounds.

pdf-tools: get the numbers out of the filing

Half of modeling is data entry from PDFs: 10-Ks, investor presentations, lender reports. The pdf skill extracts tables and text so Claude can pull a five-year income statement out of an annual report and hand it to the xlsx skill to structure. The two pair naturally, extract with one, build with the other.

cp -r ~/.claude/skills-temp/document-skills/pdf ~/.claude/skills/pdf-tools

The failure mode here is scanned PDFs. If the filing is an image rather than selectable text, OCR introduces errors, and a misread digit in a financial figure is the worst kind of silent bug. Spot-check the extracted numbers against the source for anything you'll model on.

deep-research: comps and assumptions you can defend

A model is only as credible as its assumptions. The deep-research skill runs multi-source web research to gather comparable company multiples, industry growth rates, and the data points behind your inputs, with sources attached. It turns "I think the market grows 12%" into "here are three sources, here's the range, here's what I'm assuming and why."

cp -r ~/.claude/skills-temp/deep-research ~/.claude/skills/deep-research

Treat the output as a research analyst's first draft, not a verified fact sheet. The skill is good at finding sources fast. It is not a guarantee the sources are right. Click through and confirm every figure you carry into the model. For finance work specifically, an unverified comp multiple can swing a valuation by a wide margin.

claude-api-skill: automate the updates you do every month

If you run the same model refresh monthly, the claude-api skill helps you build a script that pulls fresh data, updates the model, and flags what changed, calling Claude through the API instead of you doing it by hand each cycle. This is the most technical of the four and the one with a real cost dimension, since API usage is metered. Worth it for recurring institutional workflows, overkill for a one-off model.

A realistic workflow

Here's how the four chain together for a valuation:

1. pdf-tools extracts the target's financials from its latest filing. 2. deep-research gathers comp multiples and market growth data with sources. 3. xlsx-spreadsheet builds the three-statement model and DCF with live formulas. 4. You review every assumption, trace the formulas, and own the output.

The skills handle extraction, research, and construction. The judgment, what discount rate, which comps are truly comparable, whether the growth assumption is sane, stays yours. That's the right division of labor. If you want the broader toolkit beyond skills, our full list for finance professionals covers the standalone apps too.

Auditing a model someone else built

Skills aren't only for building. One job they're quietly good at is auditing an existing model. Hand the xlsx skill a workbook and ask it to trace dependencies, flag hardcoded values where formulas should be, find cells that break the calculation chain, or list every assumption driving the output. For anyone who has inherited a tangled model from a predecessor, this turns a day of reverse-engineering into an afternoon.

The same caution applies in reverse: the skill can tell you a formula exists and where it points, but whether the logic is correct for your business is a judgment call. Use it to map the model fast, then apply your own read to the parts that matter. A clean dependency trace is not the same as a correct model, it just makes the correct-or-not question answerable.

What these skills don't do

Worth being clear about the limits so you set them up for the right jobs. Skills won't decide your discount rate, won't tell you which comps are genuinely comparable, and won't catch a flawed assumption that happens to produce plausible numbers. They're fast hands, not a second analyst. The danger is the polished-output trap: a model that looks professional reads as trustworthy, and a wrong assumption inside a clean-looking model is harder to catch than one in a messy spreadsheet. The skill makes the work faster, which means it makes your review more important, not less.

Security: read before you install

Skills run with your permissions. A skill can execute code and touch your files, so a malicious or sloppy one is a real risk. Before installing any skill, finance or otherwise:

• Read the SKILL.md and any scripts. Know what it does before it runs. If you can't follow what a script does, don't install it.
• Prefer known sources. The official Anthropic skills at github.com/anthropics/skills are a safe baseline. For community skills, check the repo's stars, authorship, and recent activity.
• Be wary of anything that pipes to bash or hits the network unexpectedly. A spreadsheet skill has no reason to phone home.
• Keep financial data local. If a skill uploads your model or sends data to an unknown endpoint, that's a deal-breaker for confidential deal work.

A skill audit takes a few minutes and is cheaper than leaking a client's numbers.

Where these skills beat a standalone app

You might ask why bother with skills when there are dozens of finance AI apps. The answer is control and location. A skill runs on your machine, against your files, inside the environment you already trust, and you can read exactly what it does before it runs. A cloud app asks you to upload the model to someone else's server and trust a privacy policy. For confidential deal work, that difference is the whole decision. Skills also chain: the same Claude session that extracts a filing can build the model and audit it, with no copy-paste between four different web apps. That continuity is the part people underestimate until they've felt the friction of moving data between tools that don't talk to each other.

FAQ

Can Claude Code skills actually build a financial model? Yes, the xlsx skill builds Excel models with live formulas, not hardcoded values, so you get a linked, auditable model. You still set the assumptions and review the formula logic. The skill handles construction; you handle judgment.

Which skill is best for extracting data from financial filings? The pdf-tools skill, which pulls tables and text out of 10-Ks, annual reports, and investor decks. Pair it with the xlsx skill to move extracted financials straight into a structured model. Watch for OCR errors on scanned documents.

Are AI agent skills safe to use with confidential financial data? Only after you audit them. Read the SKILL.md and scripts, prefer skills from known sources like the official Anthropic repo, and reject anything that sends data to an unknown endpoint. Skills run with your permissions, so treat installation like running someone else's code.

Do I need to know how to code to use these skills? For xlsx, pdf, and deep-research, no. You install them once and prompt in plain language. The claude-api skill is more technical because it involves writing automation scripts, but the other three work without coding.

How do I install a Claude Code skill for finance work? Place the skill folder under ~/.claude/skills/ and restart Claude Code. The official skills are in the github.com/anthropics/skills repo; clone it and copy the folders you want. Claude loads the skill automatically when a task matches its description.