Paste an error to match it against the wiki. / 粘贴报错后自动匹配。
Error text / 报错原文
exit code 0 but zero stdout/stderrterminal blindnessno stdoutno stderr
What it means
The command may run, but the agent cannot observe terminal output, so it waits, repeats, or makes decisions without the actual result.
Most common causes
- Terminal integration bug in the host shell or desktop environment
- PTY output not forwarded to the agent process
- Wayland/Windows terminal wrapper mismatch
- A long-running command never flushes output
Fastest fix
- Run a tiny command like echo AG_OK to test output capture
- Restart the IDE and terminal session
- Use a different shell profile
- Write command output to a temp file and ask the agent to read it
Safe fix
- Prefer non-interactive commands for agent tasks
- Add explicit log files to test/build scripts
- Avoid commands that wait for TTY input
- Keep shell startup files minimal for IDE terminals
What not to do
- Do not let the agent assume success from exit code alone
- Do not run deploy commands until output capture works
- Do not hide errors behind silent scripts
How to confirm the fix
Run the smallest reproducible command first, then retry the agent task only after authentication, quota, and tool discovery are stable.
echo AG_OK pwd node --version
Related errors
- Agent execution terminated due to error - Google Antigravity →
- Error calling initialize: EOF - Antigravity MCP →
- Failed to exchange authorization code for token - Antigravity →
- Failed to get OAuth token: state syncing error: key not found - Antigravity →
- Stuck on Running / Generating - Google Antigravity agent not responding →
- Antigravity doesn't support MCP OAuth specifications →