-
Notifications
You must be signed in to change notification settings - Fork 5.5k
Open
Labels
bugSomething isn't workingSomething isn't workingextensionIssues related to the VS Code extensionIssues related to the VS Code extension
Description
What version of the VS Code extension are you using?
0.5.11
Which IDE are you using?
VS code
What platform is your computer?
MacOS arm M3
What steps can reproduce the bug?
Environment
- Model:
gpt-5-codex-high
- Tooling: VS Code extension (Codex integration)
- Scenario: Running project tests (
pnpm test
→ Vitest)
Problem
When triggering commands like pnpm test
inside the VS Code extension through the agent, if the underlying tool supports interactive/TTY mode (e.g., Vitest defaults to watch mode), the agent ends up inside the interactive UI.
This causes:
- The command never terminates on its own
- The agent session appears stuck waiting for keyboard input (
press h for help, q to quit
) - VS Code tasks or the agent conversation cannot continue
Steps to Reproduce
- Open a workspace with Vitest configured
- In the VS Code extension, ask the agent to run:
pnpm test
- The agent enters Vitest’s interactive watch mode instead of finishing
- The extension hangs, waiting for manual input
What is the expected behavior?
When the agent runs a test command like pnpm test inside VS Code, it should not drop into an interactive/watch mode (e.g., Vitest’s default TTY UI).
Instead, the agent should automatically execute tests in non-interactive mode so the run completes and exits cleanly.
For Vitest, that would mean running with --run.
Alternatively, the extension could expose a config option (e.g., "nonInteractive": true) so users can enforce this behavior.
What do you see instead?
No response
Additional information
No response
Metadata
Metadata
Assignees
Labels
bugSomething isn't workingSomething isn't workingextensionIssues related to the VS Code extensionIssues related to the VS Code extension