fix(desktop): salvage inflight-journal cluster — skip stale duplicate folds + bound journal persistence

[teknium1]

Salvages the Desktop inflight-journal duplicate-answer cluster — PR #84021 by @Nicolas-Formenton and PR #82832 by @egilewski — onto one green branch against current main, with contributor authorship preserved via cherry-pick.

Fixes #85308
Fixes #63047

What this combines

1. Bound inflight journal persistence (#63047) — from #82832, author @egilewski

The desktop journal synchronously read, parsed, cloned, and rewrote one aggregate localStorage value while streamed turns were repainting. Large tool results and multi-session state could block the renderer and leave the app completely unresponsive (~5 messages on the reporter's machine, Settings included).

• Store bounded recovery projections under per-session keys; migrate legacy v1 data once.
• Isolate quota and storage failures; keep the newest recoverable tail without letting oversized writes replace valid state.
• Keep bulk tool payloads and diagnostics out of the hot path.
• Adds a real Electron/CDP macOS-arm64 A/B repro harness (repro:short-session-hang) with native visibility control, renderer heartbeat, and Settings/composer/transcript checks, plus focused regressions.
• Includes the follow-up test fix (spy on the active localStorage implementation, not Storage.prototype) so the regression is observable under Node 26's jsdom fallback.

2. Skip stale journal duplicate folds (#85308) — from #84021, author @Nicolas-Formenton

The inflight-turn journal can outlive the turn it recorded (reclaim, reconnect, or restart races skip the settle that clears it). On session resume the fold then re-appended journaled assistant rows to a transcript that already held the committed replies — duplicate answers in scrambled order — and carried the stale entry's streamId onto the resumed state, which kept the journal alive (persistInFlightTurnState only clears when streamId is null) and re-folded the same tail on every open.

• journalTailAlreadyCommitted(): when no base user row matches the tail, drop the fold if every recoverable assistant row's text is already committed in the transcript (error-only rows keep the conservative append path).
• Idle-resume rule: only a genuinely running turn (keepPending: true) keeps a live stream target; an idle resume clears streamId so the journal self-heals instead of re-folding forever.

3. Sibling-site widening (maintainer follow-up)

The assistant-tail append exit path (user row persisted, no projection row) still carried the journal's streamId on a not-running resume — the same self-perpetuating-fold bug class. Applied the same keepPending gate there and pinned it with a regression assertion.

Verification

• npx vitest run src/lib/inflight-turn-journal.test.ts — 38/38 pass (includes both PRs' regression suites merged).
• node --test scripts/run-short-session-hang-repro.test.mjs — 7/7 pass.
• npx tsc --noEmit on apps/desktop — clean.
• npx eslint on all touched JS/TS files — clean.
• Both PR branches were based ~1.1k commits behind; cherry-picks applied onto current origin/main with one clean auto-merge and zero manual conflicts (the journal file has not moved; the use-session-state-cache.ts / use-session-actions call sites are unchanged).

Credit

• @egilewski — bounded persistence, repro harness, and the localStorage spy test fix (fix(desktop): bound inflight journal persistence (#63047) #82832), authorship preserved on 2 commits.
• @Nicolas-Formenton — stale-fold skip and idle-resume self-clear (fix(desktop): drop stale inflight-journal folds that duplicate answers #84021), authorship preserved on 1 commit.

Infographic

[github-actions]

૮ >ﻌ< ა ci review

_{ran on bfd423d — fix(desktop): clear the stale stream target on the assistant}

⚠️ Warnings

CI timings · View report · View job

Wall time 7m43s vs 3m24s (+127.0%). 15 job(s) slower, 19 faster, 3 unchanged.

• Python tests / Run tests slice 11/12: -46.0s
• Python tests / Run tests slice 2/12: -38.0s
• Python tests / Run tests slice 3/12: +35.0s
• JS & TS checks / apps/desktop / check:test:desktop:all: -32.0s
• Python tests / Run tests slice 8/12: -23.0s

---

OSV vulnerability scan · View job

5 known vulnerabilities found in pinned dependencies.

• CVE-2026-67213 in package-lock.json
• CVE-2026-67213 in website/package-lock.json
• CVE-2026-70608 in package-lock.json
• CVE-2026-56876 in package-lock.json
• CVE-2026-70606 in package-lock.json

How to fix:

Review the findings in the Security tab. Update the affected dependencies if a patched version is available.