Repository navigation
Description
Before submitting
- I searched existing issues and did not find a duplicate.
- I included enough detail to reproduce or investigate the problem.
Area
apps/server
Steps to reproduce
- Open a thread in T3 Code using the managed Antigravity provider (
agy.exe acp,1.5.13) infull-accessmode (interactionMode: "default",gemini-4-neonxhigh). - Ask the agent to run a parallel workflow via
run_workflow(e.g., 3 parallelagent(..., type_name="self")workers researching different topics withsearch_web). - Have one or more workflow child agents invoke
run_command(in our repro, 2 of the 3 workers calledrun_commandwithcurl.exe -sIto resolve Google search redirect URLs, while the 3rd only calledsearch_webandfinish). - Wait for the workflow to finish and inspect the workflow task log and child conversation SQLite DBs.
Expected behavior
In full-access mode, run_workflow child agents should be able to execute run_command (matching the parent Antigravity session) without hanging on hidden approval prompts—or fail cleanly if approval cannot be requested—and the workflow should complete normally.
Ideally, T3 Code should also indicate that a run_workflow background task is running rather than leaving the thread looking idle.
Actual behavior
- The 1 worker that only called
search_webandfinishcompleted cleanly in ~53s. - The 2 workers that called
run_commandnever spawned apowershell.exeorcurl.exeprocess and hung indefinitely (manage_taskstayedRUNNINGwithLast progress: 6m42s agountil manually cancelled). - Inspecting the child conversation SQLite DBs (
antigravity-acp/conversations/<childId>.db,stepstable) showed completedsearch_websteps atstatus = 3(DONE), while each stuckrun_commandstep stayed frozen atstatus = 9(waiting for user approval), with no approval prompt surfaced in T3 Code (t3_pending_requestempty) and no 60s timeout. - T3 Code showed no live progress or active indicator for
run_workflowwhile it was running/hung, so from the UI it was impossible to tell whether the workflow was executing or stuck. - Re-running the exact same 3-agent
run_workflowwhile explicitly instructing the child agents not to callrun_command(search_web+finishonly) completed cleanly in ~72s.
Related issues
- Closely related to [Bug]: Antigravity CLI detached subagents time out waiting for permissions in Full access #39029 (and open PRs fix(server): agy CLI Full access sessions run in the CLI's auto mode #39047, fix(server): Antigravity CLI subagents no longer time out in Full access #39053, fix(server): Antigravity subagents in Full access no longer time out on hidden permission prompts #40447):
agy.exe acpis spawned without--dangerously-skip-permissions(agy.exe acp --gemini-dir ... --app-data-dir antigravity-acp) and runs indefaultsession mode. While the parent session auto-approvessession/request_permission, hidden child conversations never forward permission requests to T3. Unlike detachedinvoke_subagentin [Bug]: Antigravity CLI detached subagents time out waiting for permissions in Full access #39029 (which sits instatus = 6and times out after 60s),run_workflowchild agents hang indefinitely instatus = 9with no timeout. - Also related to [Bug]: Antigravity parent thread shows indefinite "Thinking" while waiting on a subagent or background task instead of a waiting/monitoring state #40551 / [Bug]: Antigravity subagent thread renders unexpandable Thinking while running, then blank when Idle #40550 (missing UI state for Antigravity background work/subagents) and Antigravity ACP: background-task wakeups silently deny shell commands in full-access mode #40032 (Antigravity ACP background-task command permissions).
Impact
Major degradation or frequent failure
Version or commit
0.1.12-nightly.20261121.5064
Environment
Windows, T3 Code (Nightly) 0.1.12-nightly.20261121.5064, managed agy.exe 1.5.13, gemini-4-neon (xhigh), full-access
Logs or stack traces
Task log (stuck for 6m42s+ before manual cancel):
- [phase] Research
- Starting 3 parallel research agents
- [agent:self] WASI Researcher: running (2974af57-95c1-4df3-bcd3-7344378fba39)
- [agent:self] io_uring Researcher: running (22012723-e408-4be3-9b3b-c1b301854a51)
- [agent:self] SQLite WAL Researcher: done (216db0ba-7f36-4f4f-9f94-d949c6750bc4)
Child SQLite DB steps:
- 2974af57-95c1-4df3-bcd3-7344378fba39.db: idx 0..6 status = 3 (DONE), idx 7 step_type = 132 status = 9 (run_command, no child shell process spawned)
- 22012723-e408-4be3-9b3b-c1b301854a51.db: idx 0..7 status = 3 (DONE), idx 8 step_type = 132 status = 9 (run_command, no child shell process spawned)
Spawned process command line:
agy.exe acp --gemini-dir <profile> --app-data-dir antigravity-acpScreenshots, recordings, or supporting files
No response
Workaround
Explicitly instruct run_workflow child agents not to call run_command (or other tools requiring approval) and only use auto-allowed tools such as search_web, read_file, and finish.
Activity
S1ckZer commentedon Nov 21, 2026
annoying bug
juliusmarminge commentedon Nov 21, 2026
This is the same problem as #39029, so I'm closing it in favour of that one. If it is different, say how and it will be reopened.