Prompts & Context

An agent's prompt is its standing instructions — the text that sets its role,
voice, rules, and any background it should always have in mind. This is *how an agent
behaves*, as opposed to what it can do (abilities) or what it knows on demand (skills).

Prompts are built from slots — separate, named pieces (an overall system slot, an
agent-persona slot, a skills slot, an automation slot, and so on) that are assembled
into the final instructions for each message. Slots come in two layers: an **admin
base (the default for the agent, seeded from its template) and per-user overrides**
(your personal tweaks layered on top, either replacing or appending). This lets a
shared agent have one solid baseline while still being personalisable.

Around the prompt, the app automatically folds in context the agent needs that
turn: recalled memories, a list of its available tools, its active skills,
descriptions of any attached files, and any external data sources. You don't hand-write
those — they're generated and inserted for you.

Agents with the right ability can even edit their own prompt slots mid-conversation.

Related: How Agents Work: The Mental Model, Skills (On-Demand Knowledge).