feat(dojo): add the deepagents subagent-attribution demo

A demo page that renders each subagent's output as its own collapsible,
reasoning-style group: name and description from useSubagent, tool-call cards
inside the owning group, a checkmark when the subagent finishes, and an interrupt
handled inside a subagent.

Grouping is driven by `subagentRunId` on the messages via renderCustomMessages,
which is the consumer-side proof that attribution survives the whole path --
LangGraph's derived namespaces, the protocol events, the client's apply step, and
CopilotKit's rendering. Scoped to the Python-backed integrations, since the agent
is a LangGraph graph.

`useSubagent({ subagentRunId })` is plain shorthand: CopilotKit adopted the same
field name (CopilotKit/CopilotKit#5873), so no mapping is needed at the boundary.
If it ever needs an explicit mapping again, the two names have diverged.

Author: mme

apps/dojo/src/agents.ts

@@ -257,6 +257,7 @@ export const agentsIntegrations = {
         shared_state: "shared_state",
         tool_based_generative_ui: "tool_based_generative_ui",
         subgraphs: "subgraphs",
+        deepagents_subagents: "deepagents_subagents",
       },
     ),
     a2ui_dynamic_schema: new LangGraphAgent({
@@ -291,6 +292,7 @@ export const agentsIntegrations = {
         shared_state: "shared_state",
         tool_based_generative_ui: "tool_based_generative_ui",
         subgraphs: "subgraphs",
+        deepagents_subagents: "deepagents_subagents",
       },
     ),
     a2ui_fixed_schema: new LangGraphHttpAgent({

apps/dojo/src/app/[integrationId]/feature/(v2)/deepagents_subagents/README.mdx

@@ -0,0 +1,35 @@
+# Deepagents Subagents: AG-UI Subagent Attribution (Step 1)
+
+This demo runs the **deepagents** 3-subagent LangGraph agent and proves that
+**subagent attribution surfaces through AG-UI** to the frontend.
+
+## What This Demo Shows
+
+The deepagents integration dispatches work to specialized subagents. As those
+subagents emit messages, the integration stamps each message with a
+`subagentRunId` field. This demo reads that field straight off `agent.messages`
+and renders a small `⟐ {subagentRunId}` marker next to any message that carries
+one.
+
+This is intentionally **minimal** — the goal of this step is only to prove
+that the `subagentRunId` attribution makes it from the backend all the way to
+the frontend message objects. Polished UI (grouping by subagent, avatars,
+etc.) is a follow-up, not part of this demo.
+
+## How to Interact
+
+Send a prompt that requires the supervisor to delegate to its subagents, e.g.:
+
+- "Research the topic of octopus intelligence using your subagents and summarize the findings."
+
+Watch the **Message Attribution** panel on the right: every message lists its
+role, and any message produced by a subagent shows the `⟐ {subagentRunId}`
+marker.
+
+## Implementation Notes
+
+- `subagentRunId` is an AG-UI message field (see `@ag-ui/core`). At the time of
+  writing it isn't part of the CopilotKit `Message` type, so the demo reads it
+  via a cast (`(message as unknown as { subagentRunId?: string }).subagentRunId`).
+- The chat panel uses `useAgent` with `UseAgentUpdate.OnMessagesChanged` so the
+  attribution list re-renders as new messages stream in.