Test: /rubber-duck agent

[digitarald]

Refs: #319089

• anyOS
• anyOS

Complexity: 3

---

Context

The rubber duck agent is a built-in critic that gives Copilot a second opinion on its plans and code using a different AI model. It's gated behind the chat.agentHost.rubberDuck.enabled setting.

Setup

• Open Settings and enable chat.agentHost.rubberDuck.enabled
• Restart the agent host / Copilot CLI session

Test Steps

Setting gate

1. With the setting disabled (default), open Copilot CLI, type /r — verify /rubber-duck does NOT appear in completions
2. Enable chat.agentHost.rubberDuck.enabled, restart the session
3. /rubber-duck now appears

Automatic invocation

4. Ask Copilot to plan a non-trivial multi-file refactor (e.g., "Plan how to extract the logging service into its own module")
5. Observe the timeline output — verify Copilot consults the rubber duck agent and summarizes the critique (e.g., "The critique pointed out…")
6. Ask a trivial question (e.g., "What is 2+2?") — verify the rubber duck is NOT consulted

Manual invocation

7. Ask Copilot to produce a plan, then send: "Rubber duck your plan"
8. Verify structured feedback is returned with severity categories (blocking / non-blocking / suggestions)
9. Alternatively, type /rubber-duck followed by a prompt — verify critique is generated

Read-only behavior

10. Verify the rubber duck critique does not edit any files or run mutating commands — it only reads and reports