Repository navigation
UI Consistency: 1 new issue, 3 still open
New finding (comment posted)
apps/web/src/components/chat/MessagesTimeline.test.tsx:2410-2436— theworkRowhelper renders withactivityTestState.expandedleft at itsbeforeEachdefault offalse, so atone: "tool"entry on a settled run derives a collapsedwork-togglerow instead of theSimpleWorkEntryRowthe new assertions target.toContain("lucide-loader-circle")andtoContain(', stopped"')cannot match the toggle row's markup, so the running/stopped indicator tests fail rather than guarding the change.
Previously reported, still open (not re-raised inline)
apps/web/src/components/chat/MessagesTimeline.logic.ts:60-96—MAX_VISIBLE_WORK_LOG_ENTRIESis now declared, so the typecheck break is gone, butselectCollapsedWorkLogEntriesstill has no production consumer.WorkGroupSectionrenders every entry passingworkEntryIsVisibleInGroupwith no per-group cap, so the pinning policy never reaches the UI.apps/web/src/components/chat/MessagesTimeline.test.tsx:2519-2558— the collapsed-work-group cases assert a+N previous tool callsexpander and hiddenQueued shell/Blocked shellrows that do not exist in the shipped renderer.apps/web/src/components/chat/MessagesTimeline.tsx:239-244, 305-306—workingStepLabelandpendingBackgroundTasksremain on the prop contract with no reader in the component.
Resolved since the last run
activeTurnInProgressis passed again atChatView.tsx:7391, so the required-prop typecheck break is fixed.
Details
Note
Your check run agent prompt is: .macroscope/check-run-agents/ui-consistency.md
More information about how Check Run Agents work can be found in our Help Center.
Scope reviewed
Only the PR's own commits were considered (337063a, bf43aa3, 05f47ec); the merge base trails the target branch, so the rest of the range is base-branch history. In-scope changed files:
apps/web/src/session-logic.ts/session-logic.test.tsapps/web/src/components/chat/MessagesTimeline.tsxapps/web/src/components/chat/MessagesTimeline.logic.ts/MessagesTimeline.logic.test.tsapps/web/src/components/chat/MessagesTimeline.test.tsxapps/web/src/components/ChatView.tsx
Checks performed
- Shared primitives: the new running indicator uses the
Spinnerprimitive fromcomponents/ui/spinner, and becauseSpinnerspreads...propsafter its defaults, therole={undefined}/aria-label={undefined}overrides correctly suppress the primitive'srole="status"/"Loading"announcement inside a row that already has its own accessible name. No raw reconstruction of a primitive, no call-site overrides of primitive geometry, focus ring, or base colors. - Accessibility: the
, running/, stopped/, failedsuffixes extend the existing rowaria-labelcontract rather than replacing it; the removedMinusIcon"Empty" branch falls through to the null branch the row already supported. - No CSS, Tailwind, theme-variant, scroll/mask, or environment-routing surfaces are touched by this diff.
- Row derivation traced end to end (
deriveMessagesTimelineRows\u2192work/work-live/work-toggle\u2192WorkGroupSection\u2192SimpleWorkEntryRow) to confirm which fixtures reach the changed indicator code and that no per-group visible cap or expander exists on this base.
🔍 Agent activity — 23 tool calls
💭 I want to compare the current PR head against the prior run to see if earlier findings were addressed and whether anything new surfaced, so I'll start by checking the diff stat.
1. git_tools — d35c71d..05f47ec
2. git_tools — d35c71d..05f47ec
💭 The PR itself has three commits, but diffing against the merge base pulls in unrelated changes like the lockfile and CI config, which suggests the merge base predates the branch's actual rebase point rather than reflecting just the PR's own changes. So I should isolate the diff to just those three commits by comparing against their direct parent commit, restricted to the web app directory.
3. git_tools — 140b426..05f47ec
💭 Comparing against my prior findings from the earlier run, I see the missing constant has since been declared, so that issue is fixed, though I still need to verify whether the related selector has any actual production consumer by checking the timeline component's imports. I'm also revisiting whether the removed prop is still an issue. I need to check whether the props added with no consumer are still unresolved, and whether I already flagged that in a previous comment so I don't duplicate it. I also want to verify whether the test file's assertions about collapsed work log entries still match the current codebase by searching for the relevant strings and constants.
4. git_tools — selectCollapsedWorkLogEntries|previous tool call|MAX_VISIBLE_WORK_LOG_ENTRIES
5. git_tools — activeTurnInProgress|workingStepLabel|pendingBackgroundTasks
💭 Looking at the current head, the MAX_VISIBLE constant was added, partially addressing my earlier comment, but the dead-code function and the nonexistent "+2 previous tool calls" UI text still remain unaddressed. Since history was rewritten across new commits, I need to decide whether reposting is warranted given only a partial fix occurred. Cursor bot also flagged the same collapsed-groups issue as High severity on the prior commit, still unresolved. I should verify whether MessagesTimeline.tsx actually renders any expander for collapsed work groups by searching for "previous" references in the component.
6. git_tools — previous
7. git_tools — workEntryIsVisibleInGroup|groupedEntries|work-toggle|work-live
💭 I've confirmed the unused selector and constant aren't referenced in production, and that WorkGroupSection just filters visible entries without any cap, so the "+2 previous tool calls" text doesn't actually exist anywhere—meaning that test at line 2557 will fail. Let me look through WorkGroupSection fully to double-check there's no expander or cap logic I'm missing.
8. browse_code — apps/web/src/components/chat/MessagesTimeline.tsx
💭 Two prior findings still hold: the collapsed work log selector and its constant are dead code with no production consumer, and the collapsed-work-group tests in MessagesTimeline assert UI elements that don't exist, so they'll fail. I'm now checking whether the working step label and pending badge logic have similar issues. Since these three points were already flagged at a9f243a and history was later rewritten, the review threads still exist even if GitHub marks them outdated. Given the guidance against duplicate comments, I shouldn't repost these unless the code changed in direct response to them — and for the dead-code finding, only part of it changed (the constant is now defined), while the core issue remains the same. I want to diff a9f243a against the current head commit to see what's genuinely new since the last review before deciding what to flag.
9. git_tools — apps/web/src/** a9f243a8e91d7a54c5633eabbec0871bb0b61191..05f47ec
💭 Only two changes since my last review: the
activeTurnInProgressfix in ChatView is restored, and theMAX_VISIBLE_WORK_LOG_ENTRIESconstant is now defined. The dead code concern I flagged earlier still stands unchanged, and cursor also marked it High, so I'm weighing whether it needs a fresh comment given nothing there actually changed. I'm also double-checking whetheractiveTurnInProgressis still consumed anywhere — SimpleWorkEntryRow dropped its context read, so I want to confirm no other consumer relies on that field before flagging further inconsistency. There's also the deadselectCollapsedWorkLogEntriesplusMAX_VISIBLE_WORK_LOG_ENTRIESwith no production consumer, and the tests around 2500-2558 asserting collapsed-group UI that no longer exists, which should make the suite fail. Both were already flagged in the prior run and remain unresolved, so per the anti-duplicate rule I shouldn't re-post them as new comments — but since findings still exist, the check state should be failure rather than "all clear," with a summary referencing the unresolved items. I want to verify the test assertions around 2500-2560 really do fail given the current row-derivation logic before finalizing that.
10. git_tools — kind: "work-toggle"
💭 I've confirmed the shipped renderer only shows a collapsed work-toggle summary row, with no "+N previous tool calls" expander or per-group cap—so the new component tests expecting both active and finished tool text visible simultaneously don't match this behavior and will fail. Now I'm checking the running-case scenario where isWorking is false and there's no latestRun. progress entry, to see if that would still produce a work-live row, so I need to look at the branch conditions around lines 1030-1105.
11. browse_code — apps/web/src/components/chat/MessagesTimeline.logic.ts
💭 I'm confirming that with no active run, the component collapses individual tool entries into a grouped "work-toggle" summary rather than showing labels like "Active shell," which explains why those three new component tests fail. I'm also checking whether the earlier single-row test behaves similarly when there's just one grouped tool entry. The single-row lifecycle tests could face the same issue as the collapsed-group cases I already flagged, so I should check whether a work-toggle row for a single entry actually renders the spinner and the ", running" aria-label — I need to look at WorkGroupToggleTimelineRow directly to confirm.
12. browse_code — apps/web/src/components/chat/MessagesTimeline.tsx
💭 I confirmed the work-toggle row shows no spinner and no ", running" aria-label, which means the single-row tests fail too unless the test entries somehow reach a work row. Since the tests provide a structuredPayload with running status, I want to check whether workEntryIsInActiveRun could actually be derived from structuredPayload rather than latestRun.
13. git_tools — function workEntryIsInActiveRun|const workEntryIsInActiveRun
💭 With isWorking false and no active run, these single-entry rows collapse into work-toggle summaries, so the new lifecycle-indicator tests checking for the loading spinner or running/stopped states won't actually reach SimpleWorkEntryRow and would fail. I wonder if summarizeToolGroup for a lone command_execution just returns "Ran command" instead. The tests check for that text, which could accidentally pass, but the loader icon check would still fail on the running-tool test unless something's mocked. I should look at the top of the test file to see what mocks are set up around the activity state and logic module.
14. browse_code — apps/web/src/components/chat/MessagesTimeline.test.tsx
💭 Since the mock only forwards expandedWorkGroupIds when expanded defaults to false, work groups stay collapsed and render as the work-toggle summary instead. That means my new tests using tone "tool" take the group path rather than the SimpleWorkEntryRow path the existing "error" tone test relies on, so I suspect the new lifecycle tests will fail as written.
15. browse_code — apps/web/src/components/chat/MessagesTimeline.test.tsx
💭 I'm looking at an existing test that checks the "tool call failed" aria-label, tracing through the work-toggle component logic to figure out where that label actually gets generated and whether the same code path exists in the merge base. Given it's not part of the PR diff, it must already pass — so I need to check how summarizeToolGroup handles a single tool-like entry and where that aria-label string originates.
16. git_tools — Tool call failed
17. git_tools — export function summarizeToolGroup
💭 I'm tracing which component the failed-indicator aria-label test actually matches, checking whether line 2161 belongs to a shared row shell before the LiveWorkEntryTimelineRow definition at 2174. Wait, I'm reconsidering the work-toggle button's aria-label format — it seems to embed the summary text like "Glob, tool call failed," which wouldn't match the test's exact string "Tool call failed" unless summary is empty, and even then it wouldn't align cleanly. That makes me suspect there's another rendering path I haven't checked yet, so I want to look further into the else-branch logic around lines 1105-1145 to see if smaller groups render differently.
18. browse_code — apps/web/src/components/chat/MessagesTimeline.logic.ts
💭 I'm tracing through how the existing test at line 2320 would pass, checking whether the entry without an itemType counts as tool-like and whether the failure state produces the expected aria-label text. Maybe there's a different component generating that string elsewhere - perhaps a row shell renders "Tool call failed" independently. I should look at lines 2100-2175 to see what component sits there.
19. browse_code — apps/web/src/components/chat/MessagesTimeline.tsx
💭 I'm reconsidering whether that failing test actually produces a work-live row, since that requires isWorking plus an unsettled in-progress run, which doesn't match a failed status. Since this test isn't part of the diff, it must live on the base branch, so if CI passes there, a single failed tool entry must render the "Tool call failed" aria-label somewhere else in the markup that I haven't located yet. Searching further, I find that label only at line 3183, inside SimpleWorkEntryRow's tooltip trigger — meaning a plain "work" row is being rendered for this entry somehow, not a grouped one. That's odd, so I want to check whether the test sets expandedWorkGroupIds or some expansion state that would explain the fallback to SimpleWorkEntryRow, and I should look at the start of that test around line 2300 to see the full setup.
20. browse_code — apps/web/src/components/chat/MessagesTimeline.test.tsx
💭 I see the existing test explicitly sets expanded to true so the mock renders work rows, but the new tests leave expanded false by default in beforeEach, meaning they'll render collapsed toggle rows instead—so assertions expecting spinners, status labels, or the "+2 previous tool calls" text will fail since that content never renders in the collapsed state. This confirms the whole new test block in MessagesTimeline.test.tsx is written against a rendering path these fixtures never actually reach given the default collapsed state, which is a more precise and broader finding than my earlier comment that only addressed the collapsed-group cases. I want to verify this holds by checking whether SimpleWorkEntryRow only renders through WorkGroupSection for "work"-kind rows, since that would confirm collapsed tool-tone entries can't hit this path — and whether these lifecycle tests existed unflagged in earlier commits, which would make this a genuinely new, distinct issue worth raising given it's a different line and different underlying cause. The concrete fix seems clear: the new tests are missing
activityTestState.expanded = true, so they're rendering the collapsedwork-togglesummary instead of theSimpleWorkEntryRowthey're actually asserting against — this matches the pattern already used in the neighboring failed-marker test. I'm now weighing whether flagging broken tests like this actually falls within the "UI Consistency" check's intended scope, given its file-scope instructions. I'll post a fresh inline comment specifically on these new single-row lifecycle tests since I haven't flagged that particular issue before, while just noting the other previously-reported findings as still open in the summary rather than reposting them. Let me double check the running-indicator assumption though — it's possible a lone unfinished work entry gets handled differently by the row-derivation logic in a way that changes whether the test actually fails. That makes sense — for a single command, the label is just "Ran command", which happens to satisfy that test coincidentally. So the pending/waiting test passes by accident, but the running test should fail since there's no spinner in the collapsed summary, and the stopped test should fail since there's no ", stopped" aria-label present. I'm thinking the best fix is a single inline comment on theworkRowhelper, suggesting we setactivityTestState.expanded = truethere so both single-row tests pass — it's a self-contained change that touches lines already in the diff. Now I need to find the exact line numbers for that helper in the new file.
21. browse_code — apps/web/src/components/chat/MessagesTimeline.test.tsx
💭 That case still passes since no row appears without lifecycle signal. Now I'm drafting an inline suggestion for the workRow helper to set expanded state to true within it, targeting the specific lines in the code. Setting expanded state on that line doesn't break the collapsed/expanded tests, so I'll go ahead and post the review with this single inline comment, skipping re-raising earlier findings since they're unchanged. I also want to quickly confirm the Spinner primitive properly overrides its default aria-label and role props when explicitly passed as undefined.
22. browse_code — apps/web/src/components/ui/spinner.tsx
💭 Since props spread last, the undefined role and aria-label correctly override, matching the primitive's behavior. I also confirm the Loader2Icon class name aligns with what the test expects, so I'm ready to post the review.
23. modify_pr — The selectCollapsedWorkLogEntries selector plus its collapsed-group tests, and the unused workingStepLabel / pendingBackgroundTasks props, are still in the state described in the earlier threads on MessagesTimeline.logic.ts:76, MessagesTimeline.test.tsx:2557, and MessagesTimeline.tsx:243 — MAX_VISIBLE_WORK_LOG_ENTRIES now type-checks, but nothing in MessagesTimeline.tsx calls the selector and the shipped work group still renders every visible entry with no per-group cap or expander. Not re-raising those inline.
Agent Credits: 264 credits
Token usage: 40 uncached input · 2,406,258 cache-read input · 140,887 cache-write input · 16,888 output
Agent Credits may also include non-token charges from external tools such as web research.