fix(server): persist late Claude tool results after turn finalize

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}

[maslinedwin]

Fixes#6389

ClaudeAdapter.completeTurn force-completes any still-in-flight tools without a result, then clears inFlightTools. If the matching SDK tool_result arrives afterwards — the SDK result landing first, or finalize from interrupt, stream exit, session stop, or a non-steering turn handoff — handleUserMessage misses the lookup and drops the output with no warning. The persisted item.completed then has only toolName and input.

This keeps a bounded per-session map of those finalized tools, keyed by toolUseId. A late tool_result now emits the same item.updated + item.completed events as the happy path, including data.result, attributed to the original turn. It does not start a new turn or mark a completed turn as running.

Unknown toolUseId values after finalize do not throw. They log claude.tool_result.unknown-tool-use and are ignored. Map entries are removed when the late result is applied, evicted after 64 pending tools, and cleared on session teardown.

Coverage in ClaudeAdapter.test.ts:

• late tool_result after completeTurn still emits completed with data.result
• unknown toolUseId after finalize is a no-op
• happy-path tool_result before completeTurn still works

Grok / T3 Code

---

Note

Medium Risk
Touches Claude adapter turn/tool lifecycle and activity persistence; behavior is well covered by tests but ordering and duplicate-event edge cases matter for the thread timeline.

Overview
Fixes a race where Claude’s turn can finish before a tool_result arrives, leaving force-completed tools without output in storage and the UI.

Server:ClaudeAdapter keeps force-completed tools in a bounded finalizedToolsAwaitingResult map (keyed by tool use id). Late tool_result messages emit a terminal item.updated (with full data.result) on the original turn, plus content.delta when needed, and skip a second item.completed. Tool event payloads now include toolCallId. ProviderRuntimeIngestion persists the full payload for terminal item.updated events (not only the trimmed projection used for in-progress streaming).

Clients: Web and mobile work-log collapse merges tool.updated / tool.completed rows by toolCallId even when they are not adjacent, preserving the first row’s createdAt so ordering stays correct.

^{Reviewed by Cursor Bugbot for commit 16ba7a6. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Persist late Claude tool results after turn finalization and merge them in the thread feed

• Claude tool calls force-completed by completeTurn are now tracked in a bounded finalizedToolsAwaitingResult map so that tool_result messages arriving after turn finalization are matched to the correct turn.
• Late terminal tool_result events emit an item.updated with status completed/failed (attributed to the finalized turn) without re-emitting item.completed; unknown tool_use_id values are ignored with a warning.
• All tool lifecycle events now include toolCallId in their payload; terminal item.updated events persist the full payload (including result) instead of a size-limited projection.
• The thread feed (collapseDerivedWorkLogEntries in both web and mobile) merges non-adjacent late tool.updated entries into their corresponding completed rows by toolCallId, preserving the original createdAt.
• Behavioral Change: previously, all item.updated payloads were projected/truncated; terminal updates now bypass projection to retain the full result.

^{Macroscope summarized 16ba7a6.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7cdc2096-ed25-4de8-a55b-41a5374492f4

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR modifies runtime event handling for tool results across server, web, and mobile, introducing new state tracking for late tool completions. While well-tested, the coordinated behavioral changes across multiple surfaces warrant human verification.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 07f8c14. Configure here.}