Before submitting
Area
apps/server
Problem or use case
Grok 4.6 is the current Grok CLI default and regularly spawns in-process subagents (spawn_subagent, plus _x.ai/session/update subagent_spawned / subagent_finished).
In T3 those children do not show up in the Agents panel. The parent turn only gets a fat spawn_subagent tool row. Child-session updates are still dropped (V1 choice from #3156).
I hit this daily as a grok-4.6 user.
Proposed solution
Map Grok's subagent lifecycle onto the existing task.started / task.completed path Claude and Codex already use for the Agents panel. Keep child transcripts out of the parent timeline.
Why this matters
Multi-agent grok-4.6 turns are common and currently invisible next to Claude/Codex.
Smallest useful scope
Grok adapter only:
- emit
task.* from subagent_spawned / subagent_finished
- hide the bulky
spawn_subagent tool row
- no nested-thread UX, no child-transcript merge
Alternatives considered
#5420 is the earlier task-mapping request. #6410 is an open PR in this direction. This is the grok-4.6-scoped request I can take if that work does not land.
#538 (nested threads) is a larger follow-up.
Risks or tradeoffs
Need unique event stamps, per-taskId dedupe, and a name-based detector so unrelated tools are not treated as subagents.
Examples or references
Contribution
I use grok-4.6 every day and I am ready to implement this if the scope above looks right.
Before submitting
Area
apps/server
Problem or use case
Grok 4.6 is the current Grok CLI default and regularly spawns in-process subagents (
spawn_subagent, plus_x.ai/session/updatesubagent_spawned/subagent_finished).In T3 those children do not show up in the Agents panel. The parent turn only gets a fat
spawn_subagenttool row. Child-session updates are still dropped (V1 choice from #3156).I hit this daily as a grok-4.6 user.
Proposed solution
Map Grok's subagent lifecycle onto the existing
task.started/task.completedpath Claude and Codex already use for the Agents panel. Keep child transcripts out of the parent timeline.Why this matters
Multi-agent grok-4.6 turns are common and currently invisible next to Claude/Codex.
Smallest useful scope
Grok adapter only:
task.*fromsubagent_spawned/subagent_finishedspawn_subagenttool rowAlternatives considered
#5420 is the earlier task-mapping request. #6410 is an open PR in this direction. This is the grok-4.6-scoped request I can take if that work does not land.
#538 (nested threads) is a larger follow-up.
Risks or tradeoffs
Need unique event stamps, per-taskId dedupe, and a name-based detector so unrelated tools are not treated as subagents.
Examples or references
grok-4.6(default),grok-4.5Contribution
I use grok-4.6 every day and I am ready to implement this if the scope above looks right.