{% include "_ui_zoom.html" %} {% set active = 'builder' %} {% include "_menu.html" %} {% from "_menu.html" import menu_button %}

Agent Builder

Prompt, capabilities, and runtime policy

Editing

Loading agent definitions…

Capability policy

Tools, MCP servers, and skills

Choose the catalog entries this agent may use. The effective result on the right is always resolved by the backend.

Backend resolved

1 Available catalog→2 Allow + deny rules→3 Effective runtime access

Callable functions exposed to the model.

Strict allowlistPrevent profiles and implicit MCP discovery from widening this selection.

Advanced tool rules Wildcards and explicit denies

Raw allowlist one identifier per line; empty means unrestricted Explicitly disabled subtracted after allow rules

Packaged workflows and operational knowledge.

Strict skill allowlistExpose only the skills explicitly selected below.

Advanced skill rules Edit identifiers directly

Raw skill allowlist one identifier per line; empty means all discovered skills

Execution defaults

These settings shape new sessions. Existing sessions keep the model and prompt context they started with.

New sessions

01

Select the model identity and provider route used when this agent starts a session.

Model

02

Controls how the host handles tool calls after allow and deny rules are evaluated.

Permission mode

03

Choose whether first-party child-agent calls wait inline or detach as asynchronous work.

Delegation mode