A skill is a packaged procedure the assistant can follow: a set of instructions, usually paired with runnable code, for doing a specific job well. Build a discounted cash-flow model, run a regression over a dataset, price a bond, OCR a scanned PDF, scaffold a clean Word document. Instead of explaining the whole method every time, you turn on a skill once and the assistant reaches for it when your request matches.
A skill is a small self-contained folder. It holds a SKILL.md (a name, a short description of what it does and when to use it, then a markdown body of steps) plus optional reference docs and runnable scripts. The format is the open Agent Skills shape, not a PrivateMind-specific one, so skills written in that format are portable in and out.
Skills are opt-in per skill. A skill is off until you turn it on at Settings → Skills. Skills you write yourself are on for you by default.
How skills work, briefly
You don't need the internals, but one detail explains the behaviour you'll see: skills load in layers, cheapest first.
- The assistant always knows the name and one-line description of each skill you've enabled. That's the only part it carries into every conversation, and it's small. If you've enabled nothing, skills add nothing to your chats.
- When your request matches a skill, the assistant reads that skill's steps from
SKILL.md. - If the steps call for it, the assistant runs the skill's scripts and reads its deeper references as needed.
The practical upshot: enabling a skill doesn't make the assistant use it constantly. It makes the assistant aware the skill exists, so it can pick it up at the right moment and ignore it otherwise. Enabling a dozen skills you rarely use costs almost nothing per chat.
A skill runs in the same per-conversation sandbox the assistant already uses to execute code. It is saved, reusable code and instructions running where the assistant could already run code, so it grants no new reach: a skill can't touch data or systems the assistant couldn't already reach in that conversation, and it runs only inside the conversation that invoked it.
Turning a skill on
- Open Settings → Skills.
- Find the skill in the catalog and switch it on.
That's it. The skill is available from your next message. Turn it off the same way when you don't want it in the picture.
What's in the catalog
The built-in catalog covers common analytical and document work:
- Documents: create and edit Excel workbooks, Word documents, PowerPoint decks, and PDFs.
- Data and analysis: explore datasets and produce charts, run SQL over local files, transform large dataframes, run statistical inference, and apply classical machine learning.
- Finance: portfolio and risk analytics, fixed-income and derivatives pricing, portfolio optimisation, market and tail-risk measurement, and financial-statement analysis.
- OCR: pull text out of scanned documents.
- Structured writing: produce internal documents in a consistent format.
- Skill creator: scaffold a new skill from a short description (see below).
You don't have to memorise which skill does what. Turn on the ones that fit your work, describe the task in plain language, and the assistant matches your request to the right skill. These ship with the product and are kept up to date through normal releases.
Writing your own skill
Any user can author a skill. Two ways in, both at Settings → Skills:
- The form: name the skill, write a description of what it does and when to use it, and provide the steps. Good for instruction-led skills (a house style for memos, a checklist for reviewing a contract).
- A ZIP upload: package the skill as a folder and upload it. Use this when your skill ships scripts or reference files alongside the instructions.
If you're not sure how to structure one, turn on the skill creator and ask it to scaffold a skill for you. It produces the folder shape and a starting description you can edit.
If a skill needs a software library that isn't already available, ask your admin. Adding a library is a platform change. The skill itself is just instructions and code.
Sharing
Sharing is controlled by your org admin and is off by default. When your admin enables it, you can share a skill with a specific colleague, or an admin can promote a skill to the whole organisation.
A shared skill is still opt-in for the recipient: it shows up in their catalog turned off, and they choose whether to enable it. When they run it, it runs in their own sandbox, against their own conversation. Sharing a skill hands over the procedure, not access to anyone's session or data.
When the org turns one off
Every skill in the catalog is available to your organisation by default, but your admin can opt the org out of a specific skill. If a skill you expect to see isn't in your catalog, or stops appearing, your admin has likely disabled it for the org. Ask them.
Where next
- Chat: where skills get used, in the flow of a conversation.
- Files & sources: attach the documents and data a skill works on.
- Memory: another per-user feature you enable at Settings.
