Twelve interactive card types.

One protocol, twelve cards

Agents don't return free text you have to style — they return typed cards the panel already knows how to render. There are exactly twelve card types, split into interaction cards (the ones a user acts on) and presentation cards (the ways an agent shows work). No front-end coding is required to display any of them.

Interaction cards (8)

The cards a user reads and acts on — the surface of the ask → plan → approve → act loop.

• plan — the ordered steps the agent intends to take, before it acts.
• approval — a gate that pauses a sensitive step for an explicit human yes.
• progress — live status while a run executes.
• result — the outcome of a step or run, with provenance.
• comparison — options or records set side by side.
• detail — a focused, expanded view of a single item.
• context-request — the agent asking for more input before continuing.
• save-output — save a useful result back into the workflow.

Presentation cards (4)

The ways an agent shows its work — structured output rendered without custom UI.

• markdown — formatted narrative text.
• code — a syntax-aware code block.
• mermaid — a rendered diagram.
• table — structured, provenance-tagged rows and columns.

Every tool call described above is dispatched through the governance gate — permission, policy, approval, audit, and the platform-level controls are set out on the security and data-handling page. For host-specific worked examples, see the integration paths.