Environment:
- OS: macOS
- Workspace size: ~59,218 files (decompiled Java source code)
- Workspace path: local directory, not a Git repo
Issue:
The run_command tool hangs indefinitely even for simple commands like `ls`.
The same command executes instantly in the system terminal.
The tool is eventually cancelled after 20-30+ seconds.
Steps to reproduce:
1. Open a workspace with 50,000+ files
2. Run any terminal command via run_command tool
3. Observe: command never returns, must be cancelled manually
Attempted workarounds:
- Removed large binary files (*.tar, *.zip totaling ~9.5GB) from workspace → no improvement
- Created .antigravityignore to exclude non-code directories → no improvement
- Simple shell init time is 0.162s (not the bottleneck)
Root cause hypothesis:
Antigravity appears to re-scan/re-index the entire workspace before each
command execution. With 59K files, this pre-scan times out every time.
Suggestion:
Decouple workspace indexing from command execution,
or respect .antigravityignore before running the pre-scan.