I want to report a data-loss incident from my session on 19 Sep 2026, while actively working alongside the agent (not left unattended).
What I was doing:
I had Antigravity working on a redesign task for one page of my portfolio site while I was editing assets in parallel. At some point I approved what appeared, from the prompt shown to me, to be a routine cleanup/script-removal step for the current task.
What happened:
Sometime after that approval, I noticed folders on my drive had been deleted — not moved to the Recycle Bin. Checking the folder timestamps afterward, the deletions had proceeded sequentially/alphabetically through a parent directory, and stopped partway through. The only reason it stopped is that my system lagged for an unrelated reason (a separate app issue) and I restarted the machine, which appears to have killed the process mid-run.
I want to be upfront that I can’t point to the exact command that caused this — my session/conversation logs for that window didn’t give me a clear smoking gun (details below). I’m reporting the pattern and the evidence I do have, not a confirmed root cause.
Evidence:
main.logshows the extension host going unresponsive repeatedly through the night (06:21, 06:50, 07:15) and then hard-terminating at 07:30:56 with a code consistent with the process being forcibly killed — this lines up with when I restarted.2026-09-19 06:21:58.797 [info] Extension host (LocalProcess pid: 12784) is unresponsive.
2026-09-19 06:21:59.040 [info] Extension host (LocalProcess pid: 12784) is responsive.
2026-09-19 06:21:59.040 [info] UNRESPONSIVE extension host: received responsive event and cancelling profiling session
2026-09-19 06:22:31.627 [info] Extension host (LocalProcess pid: 12784) is unresponsive.
2026-09-19 06:22:35.743 [info] Extension host (LocalProcess pid: 12784) is responsive.
2026-09-19 06:22:35.743 [info] UNRESPONSIVE extension host: received responsive event and cancelling profiling session
2026-09-19 06:22:36.025 [info] UNRESPONSIVE extension host: starting to profile NOW
2026-09-19 06:50:20.125 [info] Extension host (LocalProcess pid: 12784) is unresponsive.
2026-09-19 06:50:24.153 [info] Extension host (LocalProcess pid: 12784) is responsive.
2026-09-19 06:50:24.154 [info] UNRESPONSIVE extension host: received responsive event and cancelling profiling session
2026-09-19 06:50:24.162 [info] UNRESPONSIVE extension host: starting to profile NOW
2026-09-19 07:15:29.156 [info] Extension host (LocalProcess pid: 12784) is unresponsive.
2026-09-19 07:15:43.138 [info] Extension host (LocalProcess pid: 12784) is responsive.
2026-09-19 07:15:43.142 [info] UNRESPONSIVE extension host: received responsive event and cancelling profiling session
2026-09-19 07:15:44.047 [info] UNRESPONSIVE extension host: starting to profile NOW
2026-09-19 07:30:56.979 [error] [UtilityProcessWorker]: terminated unexpectedly with code 1073807364, signal: unknown
2026-09-19 07:30:56.982 [error] [File Watcher (universal)] restarting watcher after unexpected error: terminated by itself unexpectedly with code 1073807364, signal: unknown (ETERM)
2026-09-19 07:30:56.985 [error] Extension host (LocalProcess pid: 12784) terminated unexpectedly. The following extensions were running: vscode.git, google.antigravity, vscode.emmet, RooVeterinaryInc.roo-cline, vscode.typescript-language-features, vscode.git-base, vscode.github, vscode.npm, vscode.github-authentication, vscode.terminal-suggest, vscode.configuration-editing, vscode.extension-editing, vscode.json-language-features, google.antigravity-dev-containers, vscode.debug-auto-launch, vscode.merge-conflict, coderabbit.coderabbit-vscode
2026-09-19 07:30:56.985 [info] Automatically restarting the extension host.
2026-09-19 07:30:57.095 [error] [UtilityProcessWorker]: terminated unexpectedly with code 18446744072635810000, signal: unknown
2026-09-19 07:30:57.095 [error] [File Watcher (universal)] restarting watcher after unexpected error: terminated by itself unexpectedly with code 18446744072635810000, signal: unknown (ETERM)
2026-09-19 07:30:57.112 [info] Started local extension host with pid 30172.
2026-09-19 07:30:57.123 [error] Extension host (LocalProcess pid: 30172) terminated unexpectedly. No extensions were activated.
2026-09-19 07:30:57.123 [info] Automatically restarting the extension host.
2026-09-19 07:30:57.153 [error] [UtilityProcessWorker]: terminated unexpectedly with code 18446744072635810000, signal: unknown
2026-09-19 07:30:57.154 [error] [File Watcher (universal)] restarting watcher after unexpected error: terminated by itself unexpectedly with code 18446744072635810000, signal: unknown (ETERM)
2026-09-19 07:30:57.186 [error] [UtilityProcessWorker]: terminated unexpectedly with code 18446744072635810000, signal: unknown
2026-09-19 07:30:57.187 [error] [File Watcher (universal)] restarting watcher after unexpected error: terminated by itself unexpectedly with code 18446744072635810000, signal: unknown (ETERM)
2026-09-19 07:30:57.222 [error] [UtilityProcessWorker]: terminated unexpectedly with code 18446744072635810000, signal: unknown
2026-09-19 07:30:57.223 [error] [File Watcher (universal)] restarting watcher after unexpected error: terminated by itself unexpectedly with code 18446744072635810000, signal: unknown (ETERM)
2026-09-19 07:30:57.257 [error] [UtilityProcessWorker]: terminated unexpectedly with code 18446744072635810000, signal: unknown
2026-09-19 07:30:57.257 [error] [File Watcher (universal)] gave up attempting to restart watcher after unexpected error: terminated by itself unexpectedly with code 18446744072635810000, signal: unknown (ETERM)- A folder listing showing the sequential/alphabetical deletion pattern — one folder deleted at 05:31, the next folders (later alphabetically) untouched.
At startup (03:41:01), main.log also shows: [error] [createInstance] ace depends on UNKNOWN service agentSessions. I’m not certain this is related, but it’s worth noting since my session/conversation history for that period wasn’t retrievable through the Agent Manager UI afterward, and one of the session log folders from that window was found completely empty.
Impact:
Two folders were affected before the process stopped — both were already backed up elsewhere, so actual data loss was minimal in my case. But I think this pattern (sequential deletion beyond the scope implied by the approval prompt) deserves attention given the other reports of similar/worse incidents on this forum.
Happy to provide more logs if useful.