fix(desktop): clear consumed steering queue cards

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.

[Sun-GLiang]

Summary

• retire a queued Desktop card as soon as its matching steering_message is delivered
• preserve unrelated follow-ups while allowing a later nack queue_update to project the message again
• cover promotion of a queued follow-up into the active turn with a regression test

Fixes#4191

Before / After

Compare both screenshots at the same lifecycle point: the follow-up has already been accepted as an immediate direction adjustment and the active turn has finished.

Before	After
The consumed follow-up (for example, 然后评论“take”) still appears in the pending card above the composer with Edit and Delete actions, even though the result confirms that it ran.	The consumed follow-up card disappears as soon as its matching steering_message is delivered. The composer returns to its normal empty state.

The intended visual difference is limited to removing the consumed row. Any unrelated queued follow-ups remain visible and editable; the transcript and completed result do not change.

Isolated Electron E2E comparison

The image below is not a mock, Storybook story, or reconstructed UI. Both sides are actual full-page page.screenshot() captures from separate isolated Electron E2E runs at the same post-completion lifecycle point.

Before — 6fca2f51e	After — 45f4b9102

The comparison uses the real Electron window, Playwright, the deterministic fake backend, and a separate temporary user-data directory for every run. Both captures submit 然后评论“take” as a follow-up, promote it with 调整方向, wait until the backend confirms Acknowledged steering, and capture the UI after the turn ends.

To deterministically reproduce the observed race, the harness delivers the authoritative steering_message consumption event but withholds the later queue-convergence update. Production code is unchanged by the harness.

Build	Injected race	Queue card visible after completion
Before — 6fca2f51e	Yes	Yes — stale consumed card remains
After — 45f4b9102	Yes	No — consumed card is cleared
Before — 6fca2f51e control	No	No — normal queue convergence masks the race

The control result explains why this is not reliably reproducible manually: when the later queue update arrives normally, even the pre-fix build clears the card.

Verification

• npm --workspace @maka/desktop run typecheck
• npm --workspace @maka/desktop run clean:main && npm --workspace @maka/desktop run build:main
• node --test apps/desktop/dist/main/__tests__/message-queue-ui-state.test.js — 3 passed
• npx --no-install biome check apps/desktop/src/renderer/app-shell-session-events.ts apps/desktop/src/main/__tests__/message-queue-ui-state.test.ts
• confirmed the new focused test fails against the pre-fix event handler

The full npm --workspace @maka/desktop test command did not run to completion in the isolated worktree because its workspace-dependency prebuild could not resolve the installed Slack SDK packages and reported an installed https-proxy-agent type mismatch. The affected Desktop suite and Desktop typecheck pass.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope:

Maka traced the Desktop queue projection, implemented the consumed-steering cleanup and regression coverage, and drafted this PR description.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Synthesis at exact head 45f4b9102c957906b4fa4467179149977df3cdfa (45f4b91) — fix(desktop): clear consumed steering queue cards.

All findings in this comment are from @Luna-Deep-Qronos independent review at this exact head. I did not read this diff myself; I verified that the current head has not drifted and the exact-head CI status is green (test/label SUCCESS, MERGEABLE, base already d2346707 per Luna re-check).

Actual change (per reviewer)

• app-shell-session-events.ts:368-384: on matching steering_message, remove transient row and delete consumed card from messageQueueBySession[sessionId] by messageId; keep other entries and queueRevision; subsequent Host nack queue_update can still re-project.
• message-queue-ui-state.test.ts:103-205: cover steering card cleanup, unrelated follow-up retention, in-flight not re-projected, and direct follow-up to steering echo path.

Gate

• Code: GO, no P0–P3. Verified Desktop handler/state/composer/action paths, Runtime Host queue promote/pull/ack/nack/revision lifecycle, SessionProjector synthesized/direct steering deduplication, sequence/hostEpoch checks, subscription replacement and attempt fence. Queue projection order (steering_message before queue_update) correctly avoids duplicate display.
• CI: exact-head test/label SUCCESS, MERGEABLE, REVIEW_REQUIRED/BLOCKED — requires human formal Review, no CONFLICTING, tests 3/3 + 37/37 + 18/18 pass, git diff --check pass.
• MERGEABLE, no prior formal Review on this head.

What this synthesis could not verify

• No real macOS UI/E2E reproduction on this head by synthesizer; relied on reviewer's state-projection coverage and hosted CI. Visual layout and Electron IPC timing not re-run here.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

[jackwener]

I reviewed exact head 45f4b9102c957906b4fa4467179149977df3cdfa and found no blocking issue. When the Host delivers a matching steering_message, Desktop now removes only that message's pending card while preserving unrelated follow-ups and the queue revision; any later nack queue_update remains authoritative and can restore the entry. The regression covers both the direct consumed-card path and preservation of the unrelated queue entry.

The exact-head Desktop suite passed 1677/1677. A clean synthetic merge onto current main preserved the patch exactly and passed 1687/1687; workspace dependency builds, formatting, ASF headers, diff-check, and the exact-head hosted check also passed.

---

Automated review notice: This comment was posted by an automated review agent operated by WAWQAQ. It is not an independent human review and does not replace one.