Custom Skills

You've finally written the perfect prompt. It's your earnings review, your comps layout, your house rules for a tearsheet. Don't paste it into every chat again.

Save it once as a Daloopa custom skill and run it everywhere:

• In Claude, ChatGPT, Perplexity or any other assistant connected to the Daloopa MCP.
• In Scout, Daloopa's AI agent inside Excel.

Your skills belong to your Daloopa account, not to any single chat app. A skill you save in Claude shows up in ChatGPT, in Perplexity and in Scout. Change it in one place and every place gets the new version.

Contents

1. What a custom skill is
2. Before you start: connect the Daloopa MCP
3. The four tools
4. Save your first skill
5. Run a skill
6. Edit, rename, delete and restore
7. Run the same skill in Scout
8. Compare assistants with the same skill
9. How to write a skill that travels well
10. Troubleshooting
11. FAQ

1. What a custom skill is

A custom skill is a set of instructions, written in Markdown, with a name and a short description. It tells an assistant how to do a piece of work: which data to pull, what steps to follow, which checks to run and what the output should look like.

Every skill also gets:

• an id, a UUID that identifies it;
• an enabled flag;
• created_at and updated_at timestamps.

Custom skills are private to you. No other Daloopa user can see, run or change them.

Custom skills are not Daloopa's built-in skills. The Daloopa MCP also ships prebuilt investing skills, such as DCF, comps and bull/bear, through list_skills and get_skill. Those are written by Daloopa and are the same for everyone. Your custom skills sit next to them, and you control them.

2. Before you start: connect the Daloopa MCP

Custom skills are managed through the Daloopa MCP server:

https://mcp.daloopa.com/server/mcp

If your assistant already has the Daloopa connector, you're ready. If not, follow the guide for your app on MCP Integrations. In short:

Tip: After connecting, check that every Daloopa tool is turned on in your assistant's connector settings, including the four custom skill tools below. Some apps let you switch off individual tools.

Log in with the same Daloopa account everywhere. Skills are tied to your account. If you log in to Claude with one Daloopa account and to Scout with another, they won't see each other's skills.

3. The four tools

You don't need to call these yourself. Ask in plain English and the assistant picks the right tool. Still, it helps to know what's behind the curtain.

list_user_custom_skills

Lists your skills: id, name, description, enabled, created_at, updated_at. It doesn't return the instructions themselves, so it stays fast even with many skills. Deleted skills don't appear.

Triggered by: "What skills do I have?", "Show my skills", "Run my earnings template".

get_user_custom_skill

Fetches one skill with its full instructions in content, by its id. The assistant then:

1. Reads the instructions and works out which inputs they need, such as a ticker, a peer set or a period.
2. Asks you for any input you haven't given yet. It doesn't guess.
3. Follows the instructions, pulling data from Daloopa first: fundamentals, filings, stock prices and peers.

Triggered by: "Run my earnings-snapshot skill on MSFT", "Use the skill I made for comps".

create_user_custom_skill

Saves a new skill from a name, a description and the content. The assistant shows you what it's about to save and waits for your OK. Names must be unique among your skills.

Triggered by: "Save this as a skill", "Turn what we just did into a skill called margin-bridge".

update_user_custom_skill

Changes one or more fields of a skill: name, description, content or enabled. Fields you don't mention stay as they are. Setting enabled to false deletes the skill; setting it back to true restores it.

Triggered by: "Add a valuation section to my earnings-snapshot skill", "Rename it to q-snapshot", "Delete my old comps skill".

About the id: Tools that act on one skill take its id, the UUID from list_user_custom_skills, never its name. The assistant looks it up for you. If you ask for a skill by name, expect it to list your skills first.

4. Save your first skill

Option A: describe it and let the assistant write it

Create a custom skill called earnings-snapshot. Given a ticker, pull the last 5 quarters of revenue, gross margin, operating income and diluted EPS from Daloopa. Show a table with quarters as columns, year-over-year growth for the latest quarter, and three bullets on what changed. Cite every number.

The assistant drafts the description and the Markdown instructions, shows you the draft and saves it once you confirm. You'll get back the new skill with its id.

Option B: turn a good conversation into a skill

You just spent twenty minutes getting an analysis exactly right. Lock it in:

That's exactly the format I want. Save the process we just followed as a custom skill called guidance-check, so I can run it on any ticker. Keep the ticker as an input.

Option C: paste a prompt you already have

Save this as a custom skill called bull-bear-onepager, with the description "Balanced bull and bear case for one company on one page":

(paste your Markdown here)

What makes a good first save

• Keep inputs as inputs. Say "for the given ticker", not "for AAPL", so the skill works on any company.
• Give it a specific description. It's how assistants match your request to the skill. "Earnings stuff" is vague; "One-page snapshot of the latest quarter versus the prior year" is specific.
• Use a short, lowercase, hyphenated name, such as earnings-snapshot. It reads well everywhere and is easy to type as /earnings-snapshot in Scout.

5. Run a skill

Just ask for it:

Run my earnings-snapshot skill on NVDA.

Use my comps skill for CRM, NOW and WDAY.

What custom skills do I have? Run the one for guidance tracking on AMZN.

Behind the scenes, the assistant:

1. Calls list_user_custom_skills to find the skill and its id.
2. Calls get_user_custom_skill to load the instructions.
3. Asks you for any missing input. For example, "Which periods do you want covered?"
4. Pulls the data with Daloopa's tools: get_company_fundamentals, search_documents, get_document_content, get_stock_prices and so on.
5. Builds the output the skill describes, with numbers cited back to Daloopa's source documents.

Tip: Name the skill and give the inputs in the same message ("Run earnings-snapshot on NVDA for the last 4 quarters"). The assistant can then go straight to work without asking.

6. Edit, rename, delete and restore

Talk to the assistant the way you'd talk to a colleague:

Things to know:

• Edits to the instructions replace the whole body. The assistant loads the current version, applies your change and saves the full new text. Ask to see the result if you want to check it.
• Deleting is reversible. A deleted skill is hidden, not destroyed. Restoring it needs its id; the assistant can find it for you if you still have the conversation, or you can ask support.
• Renaming keeps the id. Anything that refers to the skill by id keeps working.
• Changes apply everywhere at once. The next run in Claude, ChatGPT, Perplexity or Scout uses the new version.

7. Run the same skill in Scout

Scout is Daloopa's AI assistant inside Excel. It builds and updates financial models from plain-language requests, using Daloopa data. It reads from the same skill library as the MCP, so any skill you save from a chat assistant is ready to use in Excel.

Set up Scout

1. Install the Daloopa Add-in from Microsoft AppSource or ask your admin to deploy it. Full steps are in Installing & updating the Add-In.
2. Open it in Excel and log in with the same Daloopa account you use for the MCP.
3. Click Switch to Scout in the upper-right corner of the add-in.

Scout is part of some Daloopa plans. If the prompt box is disabled, contact [email protected].

Run your skill

In the Scout prompt box, type / and the skill's name, then the details of your request:

/earnings-snapshot MSFT, last 5 quarters, on a new sheet called Snapshot

Scout applies your skill's instructions to the workbook. For bigger jobs it shows you a plan first and writes nothing until you approve it. See Using Skills in Scout for how Scout handles skills, and Scout's built-in skills such as /dcf, /comps and /model-three-statement.

The loop: draft in chat, run in Excel

A workflow that works well:

1. Draft the process in Claude or ChatGPT, where it's quick to iterate in conversation.
2. Save it as a custom skill once the output looks right.
3. Run it in Scout with /your-skill to build it into a real model.
4. Refine it from any of them. Scout and every connected assistant pick up the change.

Writing skills that work in Excel too

A skill run in Scout ends up in a workbook, not a chat reply. To make one skill work well in both:

• Describe the output as a layout, not only as prose. For example: "a table with metrics as rows and quarters as columns, with growth rates in the row below each metric."
• Say where things go when it matters, such as "on a new sheet" or "starting at B2", or leave placement to the request so you can choose each time.
• Avoid steps only a chat can do, such as "search the web". Scout uses Daloopa data only. Keep data steps to fundamentals, filings and prices that Daloopa covers.
• Leave formatting to Scout's Global instructions if you've set them in Scout Settings (colors, units, number formats), so your skill doesn't fight them.

8. Compare assistants with the same skill

Because the instructions live in Daloopa and not in any one app, a custom skill is a fair test: same instructions, same data source, different model.

1. Connect the Daloopa MCP in two or more assistants, for example Claude, ChatGPT and Perplexity.
2. In each, send the identical request:

   Run my earnings-snapshot skill on AAPL for the last 5 quarters.

3. Compare the outputs side by side. Things worth checking:
   • Accuracy: do the numbers match? Click the citations through to Daloopa's source documents.
   • Instruction-following: did it do every step, in order, in the requested format?
   • Judgment: are the takeaways sensible and grounded in the data?
   • Effort: how many questions did it ask, and how long did it take?
4. If one assistant keeps getting a step wrong, it's often the instructions. Tighten that step, save, and rerun everywhere.

Tip: Use the same inputs and periods in every run. "Last 5 quarters" depends on the run date; say "Q3 2025 to Q3 2026" for a strict comparison.

9. How to write a skill that travels well

The same skill will be read by different models in different apps. The clearer it is, the more consistent the results.

A template

# <Skill title>

## Goal
One or two sentences: what this produces and who it's for.

## Inputs
- Ticker (required)
- Periods (optional, default: last 4 reported quarters)
- Peers (optional; if missing, find them with Daloopa's industry peers)

Ask the user for any required input that's missing. Don't guess.

## Data
Pull everything from Daloopa:
- Fundamentals: <the metrics, e.g. revenue, gross margin, diluted EPS>
- Documents: <e.g. latest 10-Q, MD&A section; earnings call remarks on guidance>
- Prices: <e.g. daily close over the period>
If Daloopa doesn't have something, say so instead of filling it in.

## Steps
1. ...
2. ...
3. ...

## Output
- Table: metrics as rows, periods as columns.
- Then: <3 bullets / a short paragraph / a chart>.
- Cite every number back to its Daloopa source.

## Checks
- Totals tie to reported figures.
- Flag any restated or adjusted figures.

Do

• Number the steps. Models follow numbered lists more reliably than paragraphs.
• Be explicit about data. Name the metrics and documents. "Revenue, gross margin and diluted EPS" beats "the key numbers".
• Name the inputs and say what to do when they're missing.
• Describe the output precisely: the table orientation, units, rounding, sections and length.
• Add checks. A short checklist at the end catches most mistakes.
• Keep it focused. One skill, one deliverable. Build a second skill instead of adding branches.

Don't

• Don't hard-code tickers or dates unless the skill is truly about one company.
• Don't depend on one app's features, such as a specific chart widget or a file upload, if you want the skill to run everywhere.
• Don't put secrets or personal data in a skill. Skills are private to you, but treat them like any document you'd save.
• Don't write "use your own knowledge" for numbers. Every connected assistant is instructed to pull data from Daloopa first; keep it that way so outputs are cited and comparable.

10. Troubleshooting

Still stuck? Email [email protected] with the skill name and the assistant you used.

12. FAQ

Can other people in my firm use my skills?
No. Custom skills are private to your Daloopa account. To share one, send a colleague the Markdown and they can save it as their own skill.

Is there a limit to how many skills I can save?
There's no fixed limit for normal use. Keep names distinct and descriptions specific so the right skill is easy to find.

What happens if I edit a skill while it's running somewhere else?
The run that has already started keeps the version it loaded. The next run uses the new version.

Can I use a custom skill through the Daloopa API instead of the MCP?
Yes, but this guide covers the MCP. The API is documented separately in Daloopa's API reference.

Will the assistant ever change or delete my skills without asking?
Assistants are instructed to show you a new skill before saving it and to confirm before deleting one. Most apps also ask your permission before running a tool that writes. Read the confirmation before you approve.

Data sourced from Daloopa.