Repository navigation
VS Code extension: every message fails with “Invalid request” (taste learning failed: Bad Request), CLI works fine #1051
Description
Summary
In the Command Code VS Code extension, every message I send is rejected with an “Invalid request” error, even a plain “hello”. Retrying and starting a new session don’t help. The Command Code CLI works correctly on the same machine with the same account, so the problem appears to be specific to the extension.
Expected Behavior
The message is sent and the agent responds, as it does in the CLI.
Actual Behavior
The chat UI shows:
Invalid request · The request was rejected as invalid. This is usually a transient glitch; try again.
Steps to reproduce the issue
- Install the Command Code extension from the VS Code Marketplace (auto-update enabled).
- Open the Command Code chat panel and start a new session.
- Send any message, e.g. hello.
- The request fails immediately with an error.
Command Code Version
0.2.3
Operating System
Windows
Terminal/IDE
vscode
Shell
No response
Session file (optional)
No response
Fix prompt (optional)
No response
Additional context
Output panel → “Command Code” channel:
2026-10-15 10:46:11.550 [info] CommandCode extension activating (node 24.21.0)…
2026-10-15 10:46:11.551 [info] IPC server started
2026-10-15 10:46:11.551 [info] [telemetry] initialized
2026-10-15 10:46:11.551 [info] CommandCode activated in 49ms
2026-10-15 10:46:36.380 [info] [webview] webview html booted
2026-10-15 10:46:43.404 [info] taste learning failed: Bad Request
Activity
[-]Every message fails with “Invalid request” (taste learning failed: Bad Request), CLI works fine[/-][+]VS Code extension: every message fails with “Invalid request” (taste learning failed: Bad Request), CLI works fine[/+]vipulgupta2048 commentedon Oct 21, 2026
Hello @sunit-salesforce would need more context on the issue. Does this happen for all models? Have you tried logging out/logging back in? Any more context you may have?
sunit-salesforce commentedon Oct 22, 2026
@vipulgupta2048
I did some more testing and can narrow it down:
Model-specific: With GPT-6 Luna (High) selected, every message still fails with the same error. After switching to DeepSeek V4 Flash, the extension works normally. I then switched back to GPT-6 Luna and it failed again, so this is reproducible and tied to that model in the extension.
I also posted about this on Reddit for more context: https://www.reddit.com/r/CommandCode/comments/1x6dcz5/command_code_vs_code_extension_returns_invalid/