Workspace Filesystem Mount & Internal Container Failures on Basic Shell Commands during Agent Sessions

Google AI Studio
1

Subject:

Workspace Filesystem Mount & Internal Container Failures on Basic Shell Commands during Agent Sessions

Description of Issue

During prolonged app-building sessions inside Google AI Studio, the underlying workspace container consistently becomes completely unstable. This degradation cascades until even the most basic shell commands and file operations fail entirely. This occurs across multiple remixed sessions and fresh chats.

Exact Error Text & Behaviors Observed

  • Top-Level Environment Errors: "An internal error occurred.", "There was an unexpected error. Finish what you were doing.", and "The request timed out. Please try again."

  • Basic Command Failures: Core agent tasks consistently fail on primitive shell operations, including:

    • Failed: running simple echo (literal command: echo "hello")

    • Failed: listing root directory contents, Failed: listing workspace files

    • Failed: killing vitest processes (literal command: pkill vitest )

    • Failed: schedule wait / waiting 5 seconds (literal command: sleep 5)

  • Filesystem Desync Pattern: The environment frequently throws "Scanned [directory] and found 0 files" across multiple known directories (e.g., apps, apps/api, src/tests) that absolutely contain project files, indicating that the container’s volume mount has unmounted or desynced mid-session.

  • Cascading Escalation: Earlier in a chat session, the agent can successfully read files and execute code. Over time, the environment degrades until new sessions fail immediately on the very first action before executing any user code.

Diagnostics Performed

The underlying codebase (ArmSphere, a React/Vite/TypeScript monorepo) has been independently built, compiled, and tested outside of the AI Studio container. The local development environment builds cleanly, and 14 out of 15 test suites pass flawlessly. This confirms the issue is entirely isolated to the stability of the AI Studio execution environment and container volume management.

Specific Questions for Engineering / Support

  1. Filesystem Syncing: Why does the container intermittently report found 0 files for populated directories? Is there an explicit command or UI action to force-refresh the container’s volume mount without losing chat history?

  2. Session Lifetimes: Are there hard session/container time limits, memory limits, or CPU thresholds that cause the background agent environment to silently crash or lock up?

  3. Orphaned Processes: Can background tasks (such as a CPU-bound infinite loop in a test runner) from a previous chat iteration persist and exhaust resources in a remixed or adjacent container workspace? If so, how can we cleanly reset the container lifecycle entirely?

2

Hi

Do you have any screenhots or logs you can share with me via dm ?