# Agent Test Page

## Overview

A sandbox for testing agent chat by selecting an agent and interacting through a streaming chat interface.

## Route

`/utilities/testing/agent`

Optional query parameter `?agent=<agentId>` pre-selects the agent on load (used when navigating from an agent's detail page).

## Page Name

"Agent Chat" for title and breadcrumb

## Page Layout

```text
[Agent Chat]                         (page title / breadcrumb)

┌─────────────────────────────────────────────────────────────┐
│  Agent                                                      │
│  [Selected agent button]               [Select agent]       │
└─────────────────────────────────────────────────────────────┘

 (no selection)          (agent selected)
┌──────────────┐         ┌──────────────────────────────────────┐
│  Select an   │   →     │  Chat panel                           │
│  agent above │         │                                       │
└──────────────┘         └──────────────────────────────────────┘
```

## Agent Selector

The selected agent is shown in the top selector button (icon + name). If no agent is selected, the page shows a placeholder prompt to select one.

## Chat Panel

Provides a chat thread for sending messages to the selected agent.

Behavior:

- Supports optional system prompt input.
- User messages are appended to the thread.
- Assistant responses stream incrementally.
- Send switches to Stop while a request is in flight.
- Clear session resets messages, system prompt, and errors.
- Enter sends and Shift+Enter inserts a newline.
- Request errors are shown inline.

## API Behavior

### `POST /api/workspaces/[workspaceId]/agents/[id]/test`

Runs a streaming chat completion request for the selected agent and returns assistant output as an event stream. Non-success responses return an error payload.