Domains

Domains are user-activated execution modes. A domain changes an agent's behavior for an explicit session and can expose extra abilities, tools, and scopes. They suit work that needs user intent, auditability, or a temporary shift in operating rules.

Additive, not replacing

The key contrast with abilities: a domain is additive. When active, its developer_prompt_addon is appended to the agent's existing developer prompt rather than replacing it. The agent stays itself and gains extra guidance and capabilities on top, for the duration of the session.

What a domain owns

• name — stable internal name
• description — what changes when the domain is active
• command — the user activation command, usually #name such as #creator
• abilities — ability refs available in this mode
• mcp_servers — MCP server refs available in this mode
• platform_scopes — permission refs. Agents must not grant or change these; users or platform-controlled setup assign them.
• prompt_config.developer_prompt_addon — extra guidance appended while the session is active

Runtime behavior

When a user activates a domain:

1. The agent receives the domain's developer prompt addon (appended to its own).
2. The agent gains the configured mode-specific abilities and tool refs.
3. A domain session is tracked for auditability.
4. The expanded behavior stays active until the user deactivates it or the session ends.

Common patterns

• #creator — write-enabled resource-building mode
• Production mode — deployment-oriented behavior with stricter audit rules
• Legal review mode — specialized review behavior and document-sensitive tools
• Debug mode — additional diagnostic behavior for a session

Abilities vs. domains

Use a domain when the work is sensitive, elevated, strategic, or should leave a clear audit trail.