Support activation welcome cards for custom Copilot CLI agents

Describe the feature or problem you'd like to solve

Problem Custom Copilot CLI agents are selected with /agent <agent-name>, but selection only changes the active profile. The CLI does not show an immediate, agent-defined welcome or onboarding message. This creates a discoverability problem for specialized workflow agents. A new user can successfully select an agent but still not know what it is for, which goals it supports, or how to begin. The user must already know an appropriate first prompt before the agent can help.

Proposed solution

Add a first-class optional welcome declaration to custom-agent metadata and
render it directly in the Copilot CLI UI after a user successfully selects an
agent with /agent.

Illustrative agent profile:

---

name: analysis-agent
description: "Helps investigate reports, inputs, and execution results."
welcome:
title: "Welcome"
message: >
This agent can help investigate existing results, analyse supplied inputs,
and prepare focused follow-up work.
suggestedPrompts:
- "Investigate an existing report or execution result"
- "Analyse a supplied input"
- "Help me understand an existing failure"

CLI behavior:

1. The user selects a custom agent with /agent <agent-id>.
2. Once selection succeeds, the CLI reads the selected agent's optional
   welcome metadata.
3. If present, the CLI renders a lightweight terminal welcome card immediately.
4. The card displays the title, message, and optional suggested prompts.
5. Selecting a suggested prompt deliberately submits that text as a normal user
   prompt. The user may instead dismiss the card and enter any free-form prompt.
6. The welcome card is UI-only:
   • it does not invoke the model;
   • it does not consume AI credits;
   • it does not create synthetic user or assistant messages;
   • it does not alter conversation history.
7. Agents without welcome metadata retain today's behavior unchanged.

Suggested terminal rendering:

┌ Welcome ───────────────────────────────────────────────┐
│ This agent can help investigate existing results, │
│ analyse supplied inputs, and prepare focused work. │
│ │
│ What would you like to do? │
│ │
│ 1. Investigate an existing report or execution result │
│ 2. Analyse a supplied input │
│ 3. Help me understand an existing failure │
│ │
│ Press a number, or Esc to enter your own prompt. │
└────────────────────────────────────────────────────────┘

Example prompts or workflows

Consider two custom agents supplied by a plugin:

• An analysis-oriented agent for investigating existing artifacts, reports,
  execution results, and input data.
• A developer-oriented agent that includes those capabilities plus source editing,
  validation, review, and delivery workflows.

After selecting the analysis agent with /agent, the CLI could immediately show:

Welcome

This agent can help investigate existing results, analyse supplied inputs, and
prepare focused follow-up work.

What would you like to do?

• Investigate an existing report or execution result
• Analyse an input or dataset
• Prepare focused follow-up work
• Understand an existing failure

The developer-oriented agent could show the same concise user goals with one short
additional note that it also supports implementation, validation, and delivery.

Requested behavior

Please add an optional custom-agent activation/welcome contract that the CLI renders
immediately after a successful interactive /agent selection.

Suggested shape (illustrative only):

welcome:
title: "Welcome"
message: "This agent can help with..."
suggestedPrompts:
- "Investigate an existing report"
- "Analyse a supplied input"

Additional context

No response

Empieza por localizar el análisis de metadatos de custom-agent y el flujo interactivo de selección de /agent en Copilot CLI. Traza cómo la selección actualiza el perfil activo y determina después dónde encajarían una tarjeta de bienvenida exclusiva de la UI y el envío de un prompt sugerido; se considera terminado cuando los metadatos de bienvenida se renderizan después de una selección correcta sin cambiar el historial ni el comportamiento de los agentes que no los tienen.