Skills
Packaged behaviours that make an agent repeatable across people and sessions.
Skills are not features. They are packaged behaviours that make an agent behave the same way for everyone who uses it. A good skill codifies how the agent should work, not just what it should output.
Skills sit alongside knowledge, which supplies grounded context, and policies, which set the boundaries.
What a skill can define
Conversation style and flow. Whether the agent is consultative and question-led, structured and checklist-driven, or direct and resolution-oriented.
Interaction instructions. What to ask, what to avoid, how to sequence follow-ups, what to do when an answer is ambiguous.
Persona in practice. Persona sets the role; skills turn it into repeatable behaviour such as discovery questioning, triage and narrowing, objection handling, or confirming requirements before moving on.
Output discipline. How summaries are formatted, how signals are structured, how drafts come out consistent between sessions.
Tool behaviour. Where tools are allowed, when to use one, what it needs first, and what to check before anything leaves the building.
The effect is that behaviour becomes a set of reusable, auditable modules rather than a fragile block of instructions.
Skills and knowledge
Skills decide how the agent works. Knowledge decides what it can draw on. The usual arrangement:
- A conversation skill sets the flow: the questions, the confirmations, when to escalate.
- A retrieval skill decides when knowledge should be consulted, such as before answering or when policy has to be checked.
- Knowledge supplies the authoritative content.
- Output skills enforce the structure of what comes out.
- Policies constrain all of it.
Skills and tools
Where tools are enabled, skills mediate them: when invoking one is appropriate given the conversation, how to read the result, and how to present what came back so it looks the same every time. That is what lets tools add capability without adding proportional risk.
Where this lives in the product
Skills are written in the console under Skills, with a name, a set of instructions, a longer document, and labels. They can be imported from a repository, attached to teams, and attached to agents in an agent's configuration. The number of active skills is capped by your plan.