When the Agent runs shell commands (non-interactive / reused PTY per Cursor behavior), commands often appear to hang indefinitely or produce no captured stdout/stderr, even for trivial commands. The same commands run normally in the user-opened integrated terminal in the same workspace.
Steps to Reproduce
Open a workspace on Windows with PowerShell as default profile.
Configure automationProfile with -NoProfile as above.
Ask Agent to run minimal commands: ver, where.exe python, echo test.
Compare output/completion with the same commands in a manually opened integrated terminal.
Hey, thanks for the detailed report. This is a known class of issues with the agent shell on Windows. There are a few overlapping tickets here like bootstrap hang, PowerShell profile not loading, and encoding on non-English locales. We don’t have an exact ETA for a fix yet.
Before we dig deeper, a few questions and steps:
Version 3.0.12 is not the latest. Newer stable builds include improvements around the agent terminal on Windows. Please update to the latest stable and check if it still happens.
What are your system locale and code page chcp in cmd? If it isn’t 65001 or UTF-8 like GBK, CP936, CP866, this might be a separate encoding case that drops stdout.
Quick workaround to test: switch your default terminal profile to Command Prompt cmd.exe and see if the agent behaves the same. If cmd works reliably, that points to the PowerShell profile or PTY layer.
If you’re using terminal.integrated.automationProfile.windows, please share the full JSON config for that block and for defaultProfile.windows too.
Related threads with the same symptom, they also include workarounds: