Test: Reasoning in Copilot CLI

[DonJayamanne]

Refs: #304641

• anyOS @joshspicer
• anyOS @aeschli

Complexity: 3

Create Issue

---

Thinking Effort Level in Model Picker for Copilot CLI

(same as #303685, but this is specific to Copilot CLI)

Enable/Disable this setting github.copilot.chat.cli.thinkingEffort.enabled by setting it to true/false
Note: This feature is on by default

A new Thinking Effort submenu has been added to the model picker for models in Copilot CLI

1. Picker Works:

   • Ensure you are using Copilot CLI agent (not Local)
   • Select a reasoning model (e.g., Claude Sonnet 4.6, GPT-5.4); verify a ">" arrow appears on the picker item
   • Click the arrow; verify a "Thinking Effort" submenu shows effort levels for that model
   • Select an effort level; verify the picker label updates (e.g., "Claude Sonnet 4.6 · High")
   • Select a non-reasoning model (e.g., GPT-4.1, GPT-4o); verify no submenu appears
   • Try a few different Anthropic and OpenAI models
   • Available levels vary per model (e.g., some have None, some skip Medium)

2. Effort Reaches the Backend:

• First locate the <home>/.copilot/session-state folder
• Rename this folder to something else e.g. session-state-backup
  This will temporarily remove all of your Copilot CLI sessions (you can just restore this folder to get all of your sessions back)
• Now start a session with a specific thinking/reasoning level
• Identify the new session created in this folder
• Look for the file events.jsonl and look for a row in this file that contains an entry similar to the following

"selectedModel":"claude-sonnet-4.6","reasoningEffort":"medium"

• As you send more requests with different model/reasoning levels, you'll get new entries in this file with different event, verify the latest entry contains the correct model and reasoning effort.

[joshspicer]

I am seeing the requests sent with the same reasoning effort in the UI. I also see the session.model_change events in the log ✅