fix(chat): smooth worktree setup and keep live activity visible

[rynfar]

Third PR of the 9b2d04317..b883fc066 batch. Three upstream commits, hand-merged — worktree setup no longer shows a jumping "Working for 0s", agent activity stays visible between actions, and a finished command stays live until the next one starts.

• #8922 smooth worktree setup status
• #8984 keep agent activity visible between actions
• #9098 keep the latest command live between messages

Skipping a commit upstream reverted

#9062 landed this behaviour and upstream reverted it 8 minutes later in #9096; #9098 is the corrected re-land. Both #9062 and its revert are skipped — only #9098 is here. The delta shows the original bug: it reused the stable LIVE_ACTIVITY_ROW_ID even when the group was not live, so the indicator vanished once a command finished.

Pylon features kept

#8922 and #9098 both rewrite the working-indicator row that carries Pylon's workingStepLabel — the same field that was silently stripped once before and restored in #221. Upstream's version of that JSX has no equivalent and would have dropped it again.

Resolved by taking upstream's structure (they remove an outer wrapper) and keeping Pylon's step-label span inside it. workingStepLabel 8 refs, activeTurnInProgress 6, reportedCostLabel 3 — all unchanged. ChatView keeps both its own props and upstream's new isPreparingWorktree; session-logic keeps Pylon's three skip conditions alongside upstream's isWorktreeSetupActivity guard.

Three test corrections, and one I got wrong

• #8922 pins the working row to h-6 so the label swap cannot change row height. Pylon's assertion predated it — updated rather than loosened.
• #8984 replaced the working row's showThinking flag with a separate thinking row. My keep-both merge left a stale assertion on a property that no longer exists.
• #9098 gives the Thinking row a brain glyph and tightens its padding to px-0.5, so the test named "aligns the iconless Thinking row" was asserting a premise that no longer holds. Renamed, and it now asserts the actual alignment contract: the size-6 icon box matching the working row's h-6.

I also briefly added a lucide-x assertion to a test whose entire point is that the icon stays neutral on a trailing failure. The suite caught it immediately and it is reverted — the failure path is still covered at three other sites via tool call failed.

Verification

@t3tools/web typecheck clean, lint 0 errors, 642 tests passing. Zero T3 Code in any added line.

Reviewed and integrated with Claude Opus 5 in Claude Code.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.8 KiB	13.6 KiB	−203 B (−1.4%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.8 KiB	6.6 KiB	−199 B (−2.9%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.7 KiB	57.2 KiB	−1.5 KiB (−2.5%)	66.4 KiB	✅
Codex	Live turn messages	11	10	−1 (−9.1%)	21	✅
Claude	Total thread wire	13.8 KiB	13.6 KiB	−171 B (−1.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.8 KiB	6.6 KiB	−173 B (−2.5%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.5 KiB	58.1 KiB	−1.5 KiB (−2.5%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: 03defac · PR result: f41b667 · Source CI: success

Scenario and decoded snapshot size

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

• Codex decoded thread snapshot: 109.5 KiB
• Claude decoded thread snapshot: 110.2 KiB

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