Regression: Sub-agents hang indefinitely on interactive terminal prompts in v0.32.0

[adamoutler]

What happened?

In version 0.32.0, when a sub-agent executes a shell command that requires user interaction (e.g., npm install, npm init), the CLI hangs silently. The sub-agent is unable to communicate the prompt to the user. This is a regression, as reverting to v0.31.0 restores the agent's ability to communicate prompts requiring user intervention.

Steps to Reproduce

1. Initialize the Gemini CLI v0.32.0
2. Deploy a sub-agent with shell execution capabilities.
3. Task the sub-agent with an action that triggers an interactive terminal prompt (e.g., installing a package that asks for funding or configuration, or running npm init).
4. The execution hangs silently.
5. Revert to Gemini CLI v0.31.0
6. Observe agent can now communicate prompts to install requiring user intervention.

What did you expect to happen?

The CLI should handle interactive shell commands initiated by sub-agents in one of the following ways:

1. Pass stdout and stdin directly to the user's terminal so the user can satisfy the prompt.
2. Intercept the prompt, pause execution, and ask the user for input via the standard CLI chat interface.
3. Timeout the command and return an error to the sub-agent so it can attempt a self-correction.

Actual Behavior

The CLI and sub-agent hang indefinitely waiting for standard input that can never be provided.

Client information

Client Information

Run gemini to enter the interactive CLI, then run the /about command.

> /about
│ CLI Version                                                    0.32.0                                                                                                                │
│ Git Commit                                                     a7eb7a310                                                                                                             │
│ Model                                                          gemini-3.1-pro-preview                                                                                                │
│ Sandbox                                                        no sandbox                                                                                                            │
│ OS                                                             linux                                                                                                                 │
│ Auth Method                                                    Logged in with Google (adamoutler@gmail.com)                                                                          │
│ Tier                                                           Gemini Code Assist in Google One AI Ultra                                                                             │
│                                                                                                             

Login information

Google Account

Anything else we need to know?

Currently, the only fix is to heavily prompt the sub-agent to strictly use non-interactive flags for all shell commands (e.g., npm install -y --no-fund --silent).

[gemini-cli]

This item has been automatically marked as stale due to 60 days of inactivity. It will be closed in 14 days if no further activity occurs. Thank you!

[gemini-cli]

A high-impact regression where sub-agents hang on interactive terminal prompts.