Root cause found for the renderer restarts in ticket e2708695 (Dev Command processes dying).
AgentsRoom runs git --no-optional-locks status --porcelain=v1 -z -uall on the project folder about every 3 seconds. It does not wait for the previous scan to finish, and it never kills a slow scan. When a single scan takes longer than the interval (here: a project folder that is a git repo with a very large untracked tree), scans pile up without limit.
Observed: more than 400 concurrent git status processes, load average 114, high tccd CPU on macOS, and the AgentsRoom main process at about 1.8 GB. Tests timed out and the renderer restarted repeatedly. Once the same command was made to finish quickly (0.08 s), there were 0 piled-up processes and no restarts for the next 11 hours, with the main process at about 250 MB.
AgentsRoom diagnostics
- App version: 1.212.0
- OS: darwin 25.6.0 (x64)
- Agent CLI: claude
- App language: th
Expected behavior
At most one status scan per repository at a time (next one starts only after the previous finishes), a timeout that kills slow scans, and a back-off or untracked-files-off fallback when scans are slow.
Actual behavior
Scans overlap without limit and never time out; the machine becomes unusable and the renderer restarts.
Steps to reproduce
Open a project whose folder is a git repository containing a very large untracked tree. Wait a few minutes, then run: pgrep -f "git --no-optional-locks status" | wc -l