I am experiencing a persistent issue with the Antigravity agent when it attempts to execute terminal commands. Even after providing the necessary approval for a command (as seen in the attached screenshot with npx prisma db push), the agent fails to execute the command and remains unresponsive.
Key observations:
Hang on Execution: After clicking approve, the progress stops at the terminal execution stage, and the agent does not proceed.
Stop Button Failure: The “Stop” button in the message input area becomes unresponsive; it does not interrupt or reset the agent’s current task.
Temporary Fix: The only way to restore functionality is to restart the application, but the issue recurs every time a new terminal command is triggered.
I have attached a screenshot showing the agent stuck during the “Loading” phase of a prisma command. Any guidance or fix for this behavior would be greatly appreciated.
The PowerShell fix works because the issue is with how bash allocates a pseudo-terminal for subprocess execution in the IDE context. Commands that expect interactive stdin (like prisma prompts for confirmation) will hang regardless - for those you can pass --accept-data-loss or equivalent non-interactive flags so the command doesn’t wait for user input.
Thanks to some great community troubleshooting, we’ve identified that terminal “hangs” after command approval are likely tied to Pseudo-Terminal (PTY) allocation in Bash. Essentially, the agent gets stuck when a command waits for a “Y/N” prompt or complex terminal features it cannot “see” or interact with.
For those who want to stay on the Bash terminal, we’d love for you to test this two-part workaround and let us know if it stabilizes your experience:
Global Shell Configuration (The “Dumb” Terminal Fix)
Where: Add this to your ~/.bash_profile (or ~/.bashrc). Why: This tells any terminal session spawned by the agent to act in a simplified “dumb” mode, stripping away complex features that cause deadlocks.
Disable aliases that might wait for terminal polling
unalias -a
fi
2. Project-Specific Wrapper (The “Prisma” Example)
Where: Keep this in your project folder at ./bin/prisma. Why: For tools like Prisma that default to interactive prompts, this wrapper forces the flags needed to bypass the “Are you sure?” questions that cause the agent to wait forever.
Bash
#!/bin/bash
Automatically append non-interactive flags
npx --yes prisma “$@” --accept-data-loss --skip-generate
Note: Once created, tell the agent: “Always use the script in ./bin/prisma for database commands.”
Does this resolve the unresponsiveness for you? By removing the “interactive questions” and simplifying the terminal mode, there should be no reason for the agent to hang. We are looking for feedback to see if this is a robust solution for everyone.
# Antigravity PTY fix - simplified terminal for agent commands
if [[ -n "$ANTIGRAVITY_AGENT" ]]; then
# Force the shell to behave as a simple pipe
export TERM=dumb
export DEBIAN_FRONTEND=noninteractive
# Disable aliases that might wait for terminal polling
unalias -a
fi
however still had issues:
run_insanity_sweep.py appeared to stall or return empty output across multiple command_status checks.
workaround: Prefixed commands with PYTHONUNBUFFERED=1. This forces Python to flush the stdout buffer immediately, providing real-time feedback in the terminal logs.
I am experiencing several issues with the Antigravity panel following the latest update. The UI freezes every time a task finishes, making it impossible to type in the message box. I frequently receive a “The window is not responding” popup.
Additionally, the overall performance has slowed down significantly, and the user experience is no longer smooth. I have also noticed that the token limit seems to have been reduced by approximately 50%.
Regarding the ‘The window is not responding’ message: I’ve made some optimizations to my project. I suspect it was originally too demanding for Antigravity to handle.
