Agents and skills

An agent is a profile for one kind of work. A skill is a reusable instruction set the agent can load when a task calls for it.

What an agent carries

Each agent has its own:

name, icon, and description;
rules and instructions;
tools and MCP servers it may use;
skills;
model and thinking level;
starter prompts;
workspace folder.

Because the configuration travels with the agent, a finance agent can see the finance folder and the ERP server while a support agent sees the help desk and nothing else.

Create and switch agents

Open Settings › Agents and create an agent.
Give it a name and instructions. Be explicit about what it should and should not do.
Pick its tools, MCP servers, skills, model, and folder.
Save, then choose it in the top bar.

You can switch agents in the middle of a conversation. The history stays; the rules, tools, and model change. This is useful when a chat starts as research and turns into a build task.

Agents export and import as .json files. Share a carefully configured agent with a colleague, or keep a team library in a repository. Review an imported agent before using it: it can point at tools, servers, and folders you did not configure.

Skills

Skills follow the Agent Skills format. You can import existing skills, write your own in the app, and invoke one explicitly with /skill:<name>.

Built-in skills include OfficeCLI and skill-creator. The agent can also choose a skill on its own when the task matches.

A skill is a folder with instructions and optional supporting files. Keep skills narrow: one job, clear steps, and examples of good output.

Slash commands

Command	What it does
`/new`	Start a fresh conversation
`/compact`	Summarize the conversation to free context
`/skill:<name>`	Invoke a skill explicitly
`/goal`	Keep working until a stated goal is done
`/research`	Run a structured research report
`/yolo`	Skip approval for tools already allowed

Type / in the message box to see what is available.