SSE routes have outputSchema: never in generated routes.ts, breaking useEventStream typing

[rblalock]

Summary

When using sse() middleware for Server-Sent Events routes, the generated src/generated/routes.ts file sets outputSchema: never for SSE routes in the SSERouteRegistry. This causes useEventStream from @agentuity/react to return data typed as never, requiring manual casting.

Reproduction Steps

1. Create an SSE route in src/api/index.ts:

import { createRouter, sse } from '@agentuity/runtime';

const api = createRouter();

api.get('/search', sse(async (c, stream) => {
  await stream.writeSSE({
    data: JSON.stringify({ type: 'token', content: 'hello' }),
  });
  stream.close();
}));

export default api;

2. Run bun run build - this generates src/generated/routes.ts

3. Check the generated SSERouteRegistry:

export interface SSERouteRegistry {
  '/api/search': {
    inputSchema: never;
    outputSchema: never;  // <-- This is the problem
    stream: false;
    params: never;
  };
}

4. Try to use useEventStream in React:

import { useEventStream } from '@agentuity/react';

interface StreamMessage {
  type: 'token' | 'complete';
  content?: string;
}

function MyComponent() {
  const { data, isConnected } = useEventStream('/api/search');
  
  // ERROR: data is typed as `never`
  // Cannot access data.type, data.content, etc.
  console.log(data.type); // TS error: Property 'type' does not exist on type 'never'
}

Current Workaround

Cast the data manually:

const { data: rawData } = useEventStream('/api/search');
const data = rawData as StreamMessage | undefined;

Expected Behavior

SSE routes should support typed output schemas, either:

1. Option A: Allow defining output type in the route that flows to the registry:

api.get('/search', sse<{ type: string; content?: string }>(async (c, stream) => {
  // ...
}));

2. Option B: Accept a generic type parameter in useEventStream that overrides the registry type:

const { data } = useEventStream<StreamMessage>('/api/search');

Environment

• @agentuity/runtime: latest
• @agentuity/react: latest
• @agentuity/cli: latest

Additional Context

This is particularly important for LLM token streaming use cases where the frontend needs to accumulate typed messages (status updates, tokens, completion signals) from the SSE stream.

[linear]

AGENT-1362