Skills
A skill is a named, reusable AI behaviour — a focused prompt packaged so it can be attached to any agent or workflow step. Build it once, use it everywhere in your workspace.
What skills are for
Skills let you extract discrete capabilities out of individual agents and share them across your workspace. Without skills, every agent that needs to summarise or classify text carries a copy of those instructions in its system prompt. With skills, you define the behaviour once and attach it wherever it’s needed.
Common patterns:
| Skill name | What it does |
|---|---|
summarise | Condenses long text into a short bullet-point summary |
extract_json | Parses structured data out of unstructured natural-language input |
classify | Labels a message as one of N predefined categories |
translate | Rewrites content in a specified target language |
sentiment | Rates text as positive, neutral, or negative with a brief explanation |
format_reply | Enforces a consistent tone and format for outbound messages |
Creating a skill
Go to Skills → New skill. Each skill has:
| Field | Description |
|---|---|
| Name | Identifier used in the UI and when listing attached skills |
| Description | Tells the agent when to use this skill — be specific |
| Instructions | The actual prompt text injected when the skill is active |
Writing a good description is the most important part. The agent uses it to decide when the skill applies. "Summarise long documents" is better than "Summariser".
Writing good instructions: treat the instructions like a focused mini system prompt. Be explicit about format, length, and any edge cases. For example, a summarise skill might say: “Produce a bullet-point summary of 3–5 points. Each point should be one sentence. Do not include opinions or interpretations.”
Attaching to agents
Open an agent → Capabilities tab → add skills. The agent’s effective instruction set is its system prompt plus the instructions of every attached skill.
The agent decides which skill’s instructions are relevant based on the conversation — you don’t need conditional logic.
Attaching to workflow steps
Workflow steps have their own skill attachments, separate from the underlying agent. This lets a single step behave differently from the same agent used elsewhere:
- An agent named “Responder” normally sends conversational replies
- When used as a “Draft email” step in a workflow, you attach a
formal_emailskill to that specific step - The agent’s core behaviour is unchanged everywhere else
Word count
Skills display a word count in the dashboard. Longer skills consume more context. If a skill’s instructions are very long, consider whether some of it belongs in the agent’s system prompt instead.