# Knowledge, skills and output

> Give an agent the sources it always consults and the skills it loads first, and choose how its replies look and where its work goes.

Under an agent's instruction, three sections shape every run: **Knowledge** is what it reads, **Skills** is how it knows to do the work, and **Output** is what it hands back.

## Add knowledge

Knowledge is what the agent always consults, like a price list, a style guide, a folder of past contracts or a website.

1. Open the agent and go to **Knowledge**.
2. Type <kbd>/</kbd> or choose **Add resource** (**+**), then pick a file, a folder, a page, a database or an app. To add a website, choose **Add website** and paste its address.
3. To add a file from your computer, paste it into the field. It's added to your Library first.

Knowledge takes files, folders, pages, data, apps and websites, up to 20 sources. To remove one, delete its chip.

## Add skills

A skill teaches Dvina how to do a kind of work, like triaging an NDA or building a proposal. An agent loads its skills first, before it starts.

1. Open the agent and go to **Skills**.
2. Type <kbd>/</kbd> or choose **Add resource** (**+**), then pick a skill.

Only skills can be added here. An agent holds up to 10.

## Get skills

Your skills live in your Feed under **Skills**. To get one:

- **From the official skills:** In your Feed, choose **Add Context**, then **Skill from Template**. Choose a skill to see **When Dvina uses it**, its **License**, its **Source** and its **Files**, then choose **Add**. It's added exactly as its maker publishes it, and your copy is yours to edit.
- **From a file:** In your Feed, choose **Add Context**, then **Skill**, and pick a SKILL.md file or a skill archive (a .zip file). An archive needs a skill folder with a SKILL.md file in it, and can hold several skills.
- **From a link:** Paste a link to a skill in a chat and ask Dvina to add it. It can be a GitHub repository, folder or SKILL.md, a skills.sh page, or a direct link to a .zip or SKILL.md file. Dvina asks for your approval first, naming the skill and where it comes from.

Either way, you confirm once before anything is added. Dvina shows where the skill comes from and, if it includes scripts, how many run in your sandbox. A skill from skills.sh whose publisher isn't an official one is marked: **It comes from a community publisher, not an official one.** When a file's skill has the same name as one of yours, Dvina says it replaces yours.

A skill whose SKILL.md has no name or description in its frontmatter stays inactive until you add them. To change a skill, open it in your Feed and edit its files.

## Choose the output

Under **Output**, choose what the agent's work looks like and where it goes.

- **Reply format:** A template every reply follows, like "Verdict, Risks, Next step". Type it in plain words.
- **Write into:** A document or a page that each run updates. Type <kbd>/</kbd> to pick it. Each run saves a new version, and earlier ones stay in its history. An agent writes into one document or page at a time: pick another, and it replaces the one before.

## Tips

- **Point to the source of truth.** Add the one file the agent must follow, like your contract playbook, instead of explaining it in the instruction.
- **Keep one running document.** Use **Write into** for work that builds up, like a weekly report, so every version is in one place.
- **Shared agents show their sources.** People you share an agent with see its knowledge, skills and output in its panel. A source they can't open shows as **No access**.

## Related

- [Create an agent](/docs/use-dvina/agents/create-an-agent)
- [Context](/docs/use-dvina/features/context)
- [Create](/docs/use-dvina/features/create)
- [Connectors overview](/docs/connectors/overview)