feat(antigravity): show subagent calls and results

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Antigravity subagent calls appeared as generic tools, so the Agents panel stayed empty. Mobile could miss their cancelled or interrupted state.

Map the official ACP start_subagent calls to existing task events. Web and desktop use the Agents panel. Mobile uses its existing task rows. Concurrent calls stay separate, results and errors remain visible, and cancellation or process exit clears active calls. Replayed calls wait for their final result before reporting success.

ACP does not expose child names, models, usage, or parent links. The UI shows only the invocation status and result. No new wire contract or agent controls.

Verified 193 focused tests for this PR and 350 focused tests with the other Antigravity changes. Server and mobile typechecks, targeted lint, and formatting pass. An isolated browser check covers two concurrent calls, success, failure, late updates, and an MCP name collision. A real Google subagent turn has not been run.

Client check

These images and the video use native-format ACP 1.1.1 events from a local fixture, not a live Google subagent session.

Before	After, running

Status change video

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Mobile work-log rules for which task.updated events are terminal changed (notably idle is only terminal with timelineBypass), and the Antigravity adapter now owns subagent lifecycle emission—incorrect classification could mis-route MCP tools or drop terminal rows.

Overview
Antigravity ACP start_subagent tool calls are no longer surfaced as generic tool items. The adapter classifies native invocations (and excludes MCP-flagged tools), emits task.progress / task.completed with per-call linkage, and tracks open subagents in session state so late merged updates cannot reopen settled calls. Open subagents are settled via task.updated when a turn ends, the user cancels or steers, or the process disconnects or stops.

Mobile work-log derivation is updated so terminal task.updated rows (completed, failed, cancelled, interrupted) can replace progress without requiring timelineBypass, task errors show as row detail, and provider status strings map to in-progress vs stopped lifecycle. Resuming agents through ordinary task.progress / task.updated no longer marks them complete incorrectly.

User docs add a Subagents section describing Agents / work-log behavior and ACP metadata limits.

^{Reviewed by Cursor Bugbot for commit faf3773. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Show Antigravity subagent calls and results as task events across platforms

• Adds native start_subagent classification and tracking in AntigravityAdapter, emitting task.progress for pending/running states and task.completed for finished states with bounded result summaries
• Settles open subagents on turn end, context shutdown, and model-change prompt cancellation by emitting terminal task.updated events with appropriate statuses (interrupted, cancelled, or failed)
• Broadens mobile work-log terminal recognition in MOBILE_TERMINAL_UPDATE_STATUSES to include completed, failed, cancelled, and interruptedtask.updated rows, not only timeline-bypassed ones; idle remains terminal only via timelineBypass
• Adds lifecycle status mappings (pending/running/waiting → inProgress, cancelled/interrupted → stopped) and task error extraction in extractWorkLogToolLifecycleStatus and toDerivedWorkLogEntry
• Behavioral Change: mobile deriveWorkLogEntries now emits terminal rows for task.updated events from providers that do not set timelineBypass; late updates for already-settled subagents in ToolCallUpdated are silently ignored

^{Macroscope summarized faf3773.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.7 KiB	13.7 KiB	+14 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+8 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	58.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−142 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−1 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−141 B (−2.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.9 KiB	−1.4 KiB (−2.4%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: f25e442 · PR result: faf3773 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 4d7945e. Configure here.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR adds a user-facing Antigravity subagent activity capability with new server-side lifecycle tracking and cross-client rendering behavior. The production scope extends beyond a bounded opt-in tweak, so the runtime and lifecycle changes merit human review.

^{You can add or adjust custom eligibility rules. Learn more.}