fix(cli): disable follow-up suggestions by default#2954
Conversation
Most Qwen OAuth users don't have a fast model configured for this feature, so it fires a wasted API request on every turn with no visible benefit. Default to off; users can opt in via settings.
📋 Review SummaryThis PR changes the default value of 🔍 General Feedback
🎯 Specific Feedback🔵 Low
✅ Highlights
|
Code Coverage Summary
CLI Package - Full Text ReportCore Package - Full Text ReportFor detailed HTML reports, please see the 'coverage-reports-22.x-ubuntu-latest' artifact from the main CI run. |
|
Disabling follow-up suggestions by default will break the follow-up tips feature in PR #2932. The We need additional documentation and tips to tell users how to enable this, for example showing a hint at the end of |
|
As stated in the PR, the Qwen OAuth users (vast majority) spend extra requests on it, but seldomly see the follow-up suggestion. I agree it is a cool feature, but right now we cannot bring equally cool experience due to external limitations. That’s why we have to opt out by default. As long as we have ability to serve a slick experience, we can always bring it back. |
|
@wenshao @tanzhenxin Let's see if we can add fast-model support to Qwen OAuth. Many upcoming features will likely depend on it, so I'll look into it — but we can keep it disabled by default for now. |
TLDR
Disable follow-up suggestions by default. Most Qwen OAuth users don't have a fast model configured for this feature, so it fires a wasted API request on every turn with no visible benefit. Users can still opt in via
ui.enableFollowupSuggestions: truein settings.Screenshots / Video Demo
N/A — no user-facing change for the majority of users (feature was already invisible to them since suggestions never rendered without a fast model). Users who had it working and want to keep it can flip the setting.
Dive Deeper
The follow-up suggestion feature generates a context-aware prompt suggestion after each turn by making an additional API call. For users on Qwen OAuth (the majority), there is no dedicated fast model for this, so the request either fails silently or returns a low-quality result that is never shown. This wastes one API request per turn.
Changes:
packages/cli/src/config/settingsSchema.ts— defaultfalsepackages/cli/src/ui/AppContainer.tsx— flip runtime check from!== falseto=== truepackages/vscode-ide-companion/schemas/settings.schema.json— defaultfalseReviewer Test Plan
ui.enableFollowupSuggestionsin settings): run a multi-turn session, check API logs — no suggestion requests should appear"ui": {"enableFollowupSuggestions": true}in~/.qwen/settings.json: run a session — suggestions should still work as beforeTesting Matrix
Linked issues / bugs