fix(ui): tighten grouped activity spacing

[Zerlight]

Summary

• tighten spacing only for activities nested inside expanded activity groups
• preserve normal spacing for group headers and top-level activities
• cover standard, tooltip-backed, and bodyless activity rows in the bundled browser smoke

Why

Grouped activity rows used the same block padding as top-level rows, which weakened the visual hierarchy. The group now scopes denser padding to each direct child header.

Validation

• pnpm check:ci
• pnpm test --maxWorkers=2 --maxConcurrency=2
• pnpm -F @linkcode/webview run e2e:browser

Linear: CODE-376

[linear-code]

CODE-376

[Copilot]

Pull request overview

Tightens the visual hierarchy of chat activity runs by reducing padding for tool/thought rows rendered inside an expanded activity run, while leaving the run header and non-grouped activity rows unchanged. It also expands the mock showcase and bundled webview smoke test to exercise tooltip-backed and bodyless activity rows under the new spacing.

Changes:

• Scope denser vertical padding to the first child header of each collapsible activity row inside an expanded ActivityRun.
• Extend the showcase mock activity run with a “bodyless” read tool call entry.
• Add a bundled browser-smoke assertion that expanded activity-run children render with denser padding than the run header, covering tooltip-trigger and non-tooltip rows.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
packages/presentation/ui/src/chat/activity-run.tsx	Applies a scoped Tailwind arbitrary selector so nested activity rows render with tighter vertical padding inside expanded activity runs.
packages/client/workbench/src/mock/data/showcase.ts	Adds a bodyless read tool-call fixture used to exercise non-tooltip activity rows in the showcase thread.
apps/webview/e2e/browser-smoke.e2e.mts	Adds a smoke check that expands the showcase activity run and validates nested row hierarchy/padding (including tooltip-trigger + bodyless rows).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[pullfrog]

ℹ️ One assertion-message nit inline; the change itself verified out cleanly.

Reviewed changes — a one-line spacing fix in ActivityRun plus the mock row and browser-smoke check that pin it. This PR is stacked on ruocheng/code-576, so the diff is against that branch, not master.

• Denser grouped rows — activity-run.tsx:157 extends the expanded-group body class with [&>[data-slot=collapsible]>*:first-child]:py-0.5, dropping each direct child header from py-1 (4px) to py-0.5 (2px) while the group trigger and top-level rows keep py-1.
• Bodyless showcase row — showcase.ts adds a read call with no content, rawInput, or locations, so hasToolBody() is false and its header renders as a plain div instead of a trigger button.
• Browser-smoke hierarchy check — verifyActivityRunHierarchy expands the mixed showcase run and asserts every child header's padding-block-start is non-zero and smaller than the group header's.

What I verified rather than assumed:

• The arbitrary variant actually compiles and wins the cascade. Running the candidate through @tailwindcss/node (v4.3.3) emits … > [data-slot=collapsible] > :first-child { padding-block: calc(var(--spacing) * 0.5) } — specificity (0,3,0) against .py-1's (0,1,0), and emitted after it in the same layer. @source "./" in packages/presentation/ui/src/styles.css covers the file, so the class is scanned.
• All three header shapes are matched. Every branch of the run.items.map bottoms out in a coss-ui Collapsible root whose first element child is the header: ToolCallItem/QuestionCallItem via Tool, ThoughtBlock via Reasoning. The tooltip-backed case is base-ui's TooltipTrigger render={…}, which merges onto the same node rather than wrapping it, so *:first-child still lands on the real header.
• The new check is CI-enforced, not local-only — ci.yml:215 runs pnpm -F @linkcode/webview e2e:browser unconditionally on pull requests — and it genuinely fails without the fix, since the children would otherwise measure equal to the group header.
• The mock addition doesn't disturb the one suite that asserts against the showcase session: dev-mock-transport.test.ts passes (13/13) with the new tool call. Resulting row height is 24px (text-sm's 20px line box + 2×2px), exactly the WCAG 2.2 AA target minimum, and no theme overrides --text-sm--line-height or --spacing.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes — b3c819f is test-only hardening of the browser-smoke check added in 8f31665; the ActivityRun spacing fix and the showcase mock row are unchanged since the last review.

• Loosened the activity-run locator — the exact accessible name 'Activity details: An action failed · Ran a command · Made a file change · Explored 2 times' is replaced by RE_ACTIVITY_RUN_DETAILS (browser-smoke.e2e.mts:25-26), so the check no longer breaks when a clause count shifts.
• Added failure messages to the two bare asserts — the tooltip-trigger and bodyless-row checks now dump metrics.children, matching their neighbours.

The one thing worth verifying about the looser locator is whether it can now match a second button and trip Playwright's strict mode. It can't — I enumerated every activity run the showcase thread renders and derived each aria-label:

• Bursts are separated by agent-message-chunk narration (dev-mock-host.ts:1391-1428), and a non-tool item flushes the in-progress run (activity-groups.ts:85-88), so each burst becomes its own run.
• explore → Explored 4 times; files → Made 6 file changes; commands → An action failed · Ran 3 commands; afterTask → Made a file change · Explored once. None carries all four clauses the regex requires.
• Only activityRun.beforeNarration matches, and only its first five items: the preceding agent-thought-chunk (dev-mock-host.ts:1416) joins as ACTIVITY_RUN_GLUE_KEY glue while mock-activity-integration classifies as brand:linear and flushes the run (activity-groups.ts:44-46, 90-91). Reasoning + read + read-bodyless + execute-failed + edit — exactly the 5 children children.length >= 5 expects.

Dropping Explored 2 times from the locator doesn't cost coverage either: the bodyless row that clause stood for is pinned directly by the tagName === 'DIV' assertion, which is now the one that reports what went missing.

  ^{｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.