[incognitojam]

When a server restart cancels background work that outlived a finished turn (for example a dev server the agent started), upstream's restart recovery prompts the agent with a note in a new run linked to the finished one through restartContinuationOfRunId. The fork's restart-continuation folding merged every linked run into one response. The note's reply became the response's final answer, and the real answer was hidden in the "Worked for" fold.

Only a continuation of a run the restart cut short now joins that run's response. A continuation of a completed or waiting run starts its own response under the "Server restarted" note, the same way a background wake starts one under its notification. The earlier answer stays visible as its own response's final message, and the note reply gets its own fold. This matches upstream, which already shows these continuations as a separate notice and response (pingdotgg#17723).

• restartContinuationResumesResponse in client-runtime decides from the source run's status. Recovery only continues a completed or waiting source when its turn had settled, so these statuses identify that case.
• Web (MessagesTimeline.logic.ts) and mobile (threadActivity.ts) stop following a link to a settled source when grouping responses, and treat that note as a response boundary.
• Updated the server-restart-continuation-note ledger invariant.

Not changed: recovery still marks some settled waiting runs it cannot checkpoint as cancelled. The client cannot tell those from a run cut mid-turn, so they still merge as before. The working timer also still carries the earlier run's work into the note's run, as upstream does for background wakes.

Validation

• In the web dev client, seeded with real threads where a restart cancelled a dev server after the agent had finished, I checked "Show deployments in PR panel" (one finished response followed by two restart notes).
  • Before: one "Worked for 15m 37s" fold whose visible answer is the note reply, "The cancelled task was only the dev preview server…". The real answer is hidden in the fold.
  • After: "Worked for 14m 45s", then the original answer and its changed-files card. Below that, "Server restarted" opens a new response, "Worked for 39s", with the note reply under it. The second note and its reply follow as another response. No page errors.
• New web and mobile tests: a finished response followed by a restart note keeps its answer and fold, and the note starts a second response with its own fold. A chain through a settled run stops at that run. The existing tests for a restart that cut a response short still pass.
• Not verified: the mobile change in a native client. The mobile feed logic is covered by the new threadActivity test. I also didn't verify a live restart that cuts a response short; that path is unchanged, and its existing tests pass.

Before	After

---

Written by an agent (Claude Code, claude-opus-5-5).

[github-actions]

Upstream divergence

Both sides are compared with upstream at the in-order tip 54b6b667f3. Excludes .repos/ and pnpm-lock.yaml.

Measure	Base	This PR	Change
Upstream files the fork changes	1,412	1,412	0
Lines changed in them	67,777	68,018	+241
Upstream files the fork removes	41	41	0
Fork-only files	473	473	0
Fork-only lines	83,034	83,034	0

Upstream files whose difference changes (7)

File	Change	Now differs by
apps/web/src/components/chat/MessagesTimeline.logic.test.ts	+92	724 lines
apps/mobile/src/lib/threadActivity.test.ts	+72	366 lines
apps/web/src/components/chat/MessagesTimeline.logic.ts	+28	443 lines
apps/mobile/src/lib/threadActivity.ts	+15	268 lines
apps/mobile/src/features/threads/ThreadFeed.tsx	+12	291 lines
packages/client-runtime/src/work-log/presentation.ts	+12	256 lines
apps/web/src/components/chat/MessagesTimeline.tsx	+10	813 lines

[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	4.9 KiB	5.0 KiB	+23 B (+0.5%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	+23 B (+1.9%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.8 KiB	20.9 KiB	+41 B (+0.2%)	29.3 KiB	✅
Codex	Live turn messages	1	2	+1 (+100.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	+24 B (+0.5%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	+24 B (+2.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	+41 B (+0.2%)	29.3 KiB	✅
Claude	Live turn messages	1	2	+1 (+100.0%)	8	✅

Baseline: 0b80358 · PR result: d0dfa1e · 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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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