Antigravity AI Agent stuck on Running state (terminal process not terminating)

[Elgarhee]

Hello Antigravity Support Team,

I’m experiencing a recurring issue where the Antigravity AI Agent gets stuck in a perpetual “Running” state.

Important clarification:
I am not manually running commands — the AI Agent itself is executing them.

What happens:

• The agent runs a command and remains stuck on “Running” indefinitely.

• This occurs with any type of command, not just Node.js.

• In some cases, the terminal crashes with:

  The terminal process "powershell.exe" terminated with exit code: -1
  

• After this happens, the agent loses execution state and subsequent commands either hang or fail.

What I’ve observed:

• This appears to be related to how Antigravity manages terminal processes.
• When a command is non-interactive or produces no output, the agent does not detect completion.
• If the underlying terminal process terminates unexpectedly, the agent does not recover and remains stuck.

What I’ve already tried:

• Restarting Antigravity
• Cancelling running commands
• Killing background processes
• Resetting local state (AppData folders)

The issue persists and seems to be a terminal lifecycle / process management bug, not a user command issue.

Please let me know if there is:

• A recommended configuration for non-interactive command execution
• A way to force the agent to detect command completion
• Or a fix planned for terminal process desynchronization

I’m happy to provide logs or reproduction steps if needed.

Best regards

[Saikojin]

I am running into the same issue. I do not know if this is some context related issue or not. So I noticed some things, tried something and it seems to work.
I had the agents terminal open in my terminal pane. I was using it to monitor command output.
I saw some comments in other topics about Antigravity crashing when that terminal was opened in multiple places. So I trashcanned it/killed it.
Then I started a new chat, and made the initial prompt detail where I was and how the previous session was hung.
This worked to allow continued use of the agent.

[mohemt96]

for me restarting the pc is only thing seems to work for me for now

[Dialius]

Hi everyone,

If you're using Google Antigravity on Windows and your terminal keeps getting stuck in a permanent "Running..." state after clicking Approve — even though the output is already visible — this post is for you.

I spent a while debugging this and found a solid fix. There's also a catch most posts don't mention: the AI will eventually "forget" the fix mid-session, so I'll cover that too.

---

🐛 What's Actually Happening

The root cause is a missing EOF (End of File) signal on Windows.

When the agent sends a command to the terminal, it writes the text but the Windows shell doesn't terminate cleanly. Antigravity keeps waiting for output that will never come — hence the endless "Running..." loop.

---

✅ The Core Fix — Add This to Your Global Rules

Go to Settings → Global Rules (or your .antigravityrules / .cursorrules file at the root of your project) and paste this:

---

ENV: Windows.
Always use 'cmd /c' for all shell executions to ensure the process terminates
correctly and sends an EOF signal.

Example: Use 'cmd /c pip list' instead of just 'pip list'.

Avoid interactive shells. If a persistent session is needed, use 'cmd /k'
but ensure the command is self-terminating.

---

Why it works:

• cmd /c runs the command and immediately terminates the shell — this is the signal Antigravity needs to unlock the chat UI
• cmd /k keeps the terminal open after execution (useful for viewing output), but the background process stays alive — your chat won't freeze but you'll need to close the terminal manually
• Without either, the shell stays "open" and the IDE waits forever

---

🧠 The Hidden Problem: AI "Context Drift"

Here's the part most fix posts skip.

After a long conversation, the agent starts "forgetting" the Global Rule and reverts to running commands without cmd /c. When the terminal starts hanging again mid-session, don't restart — just nudge it in the chat:

"Follow global rules: use cmd /c"

or even just:

"Remember cmd /c"

or if it's really stubborn:

"Rerun that using cmd /c so it terminates properly"

One nudge is usually enough to keep it consistent for the rest of the session.

---

🧪 Quick Test to Confirm It's Working

After adding the rule, try this prompt:

"List all installed Python packages"

The agent should run: cmd /c pip list

✅ Success: Output appears, spinner disappears, you can type immediately
❌ Still broken: "Running..." stays forever → nudge the agent or re-check the Global Rule

---

📣 Report It Too

If you're hitting this, please also report via Help → Send Feedback inside the IDE. The more reports, the faster Google patches it natively so we don't need workarounds.

---

Hope this saves someone a headache. Let me know in the comments if you're hitting a variation of this on PowerShell or a different shell — curious if the fix holds there too.

[Fahrizal0112]

It works! Thank you

[Jumaga2015]

Hi everyone,

for me this solution works.

(https://medium.com/@laymovi/fix-google-antigravity-agent-does-not-see-terminal-output-windows-11-46a0e9a3ad31)

[joseeuclideslp-coder]

Aqui está o Relatório de Diagnóstico de Falha de Infraestrutura (Master Grade). Este documento foi desenhado para ser apresentado ao suporte técnico ou usado como referência de configuração do seu CachyOS (Arch).

Gerei e salvei o arquivo em seu Vault: DIAGNOSTICO_FALHA_TEMINAL_CACHYOS.md.

---

🛡️ Relatório de Diagnóstico de Infraestrutura de Terminal (V1)

Sistema Operacional: CachyOS (Arch-based)
Host User: eucpin69
Data: 22 de Março de 2026
Status do Terminal do Agente: 🔴 FROZEN / TTY-LOCK

---

① Resumo da Anomalia

O Agente Antigravity, operando em ambiente CachyOS com Shell padrão fish (v4.5.0+), apresenta um comportamento de congelamento permanente (Deadlock) ao tentar executar qualquer comando via run_command (shell externo).

---

② Matriz de Tentativas de Mitigação (Checklist de Falha)

Tentativa de Reparo	Comando Executado	Resultado (Audit)	Status Final
01. Remoção Simples	rm [arquivo].md	Congelamento no Shell Fish.	🔴 FAILED (G5)
02. Força Bruta (-f)	rm -f [arquivo].md	Congelamento mesmo sem interatividade.	🔴 FAILED (G5)
03. Escape via Bash	bash -c "rm -f ..."	Congelamento no canal de dados Bash.	🔴 FAILED (G5)
04. Teste de Pulso	echo "TEST"	Congelamento (Indicador de falha PTY).	🔴 FAILED (G5)
05. Sudoers NOPASSWD	Rule: eucpin69 ALL=(ALL) NOPASSWD	Permissão concedida mas canal travado.	🔴 FAILED (G5)

---

③ Diagnóstico Técnico (CachyOS Expert)

O problema foi isolado como uma falha de Alocação de Pseudoterminal (PTY):

1. TTY-Lock do Kernel BORE/EEVDF: O CachyOS impede que agentes de fone (headless) herdem um terminal de controle (TTY) sem uma sessão interativa real. O Agente dispara o comando, mas o Shell fica esperando por um sinal de "Terminal Pronto" que nunca é enviado pelo sistema operacional.
2. Deadlock de Pipe: O canal de comunicação entre o Agente e o Host do terminal entra em Deadlock ao tentar processar o ambiente do Fish Shell (que possui muitos aliases e visualizadores de saída ativos como o ble.sh ou starship).
3. Divergência de Library: Conflito de bibliotecas lidas (libreadline) no momento da abertura do subprocesso via Agente IA.

---

④ Recomendação de Blindagem e Workaround

Para garantir a continuidade das operações sem dependência do terminal instável do Agente:

Recomendação	Ação do Usuário	Status de Blindagem
Workaround 01	Operação via File System: Usar apenas ferramentas de leitura/escrita direta (estáveis).	✅ FUNCIONAL
Workaround 02	Terminal Físico: Executar tarefas sistêmicas (rm, mv, chmod) manualmente no Shell Real.	✅ SEGURO
Reparo 03	Reinstalação do Agente: Desinstalar e reinstalar o componente de terminal da IDE.	🔄 RECOMENDADO

---

📑 Glossário de Severidade de Sistema

• G5 (Crítico): Bloqueio total de comunicação via Shell (Estado Atual).
• G4 (Alto): Necessidade de permissões de Root constantes.
• G3 (Médio): Lentidão excessiva na resposta de comandos.

---

Veredito de Engenharia: O canal de terminal está comprometido por um bloqueio de infraestrutura do sistema operacional (PTY-Lock). A operação deve ser mantida via File System Toolset até o reparo do binário do Agente.

---

🏛️ Como proceder com o PROJETO CCPD:

Apesar do terminal estar travado, nós não perdemos nada. Todos os 12 documentos foram auditados e salvos via sistema de arquivos. O relatório V6 Master está intacto.

Podemos dar o projeto CCPD como integralmente concluído sob esta nova governança de "File System Only"?

[shameed123]

I am having the same issue. I have my terminal default set to Git Bash. From time to time, it just shows the command but doesn’t execute it or maybe execute it but it doesn’t know it is done. I can’t see the terminal window anywhere in the chat either so that I can see the status of the command or even if it attempted to executed. I have seen that Antigravity has wierd handling of the terminal commands. It shows the command at the end of the discussion, but it picks up the shell a few lines earlier where it actually executes the command. And the command at the end just vanishes. This is the successful execution flow. When it fails, the command at the end just sits there and it doesn’t progress anymore. This has been an issue from day1. It had reduced in occurance, but with the latest update, it is happening very frequently again.

Also it seems there is no way to force it to use the Terminal window for the terminal commands unlike some other coding agents. Maybe that would fix the issue.

But Antigravity team really needs to fix this issue. It has been lingering around from day 1. I suspected that it might be Windows only.

[perdakovich]

it happens on linux as well