VPS2 durable per-view state - independent-fix series (supersedes #34 + the 21-item upstream draft series)

**Tracking issue for the vps2 durable per-view state series.** Replaces fork PR #34 and the closed upstream series: PRs Zoo-Code-Org#966 #928 #979 #981 #983 and issues Zoo-Code-Org#967 #978 #980 #982 #984 #913 #915 #916 #917 #918 #919 #920 #921 #922 #923 #924 #929 #1065 — 23 of the 24 closed as superseded on 2026-09-05, each with a tracker-pointer comment naming the covering fix unit (or the out-of-scope reason); upstream #1065 carries the pointer comment but its close is permission-blocked (the account is not a repo admin, 403) — flagged for a maintainer.

- **Base of record**: `upstream/main` @ `0d937c050810ebc305b34368bf6d6ae8c1f89966` — **every** fix PR bases on this commit (independence guarantee; stale sibling heads are porting sources only, never bases).
- **Main drift (2026-09-06)**: upstream/main advanced from the base of record to `8d296deef` (9 commits; v3.82.0 — #1533 release, #1319 task-history reopen, #1470 subtask-resume stall, #1525, #1527 teardown, #1534 node24 CI, #1488/#1506 providers). Porting-surface deltas: `ClineProvider.ts` — v3.82.0 announcement id (~L309); new `drainTaskDisposal` + rewritten `dispose()` that now evicts and disposes every registry task (~L812–860); `reopenParentFromDelegation` delegation-reopen fixes (~L4065–4272). `Task.ts` — 23 hunks (resume/eviction/teardown). `ClineProvider.spec.ts` — +127 new tests (~L1140) + 1 new Task-mock member (~L266). Both shipped PRs remain MERGEABLE, all CI green. The stack keeps the base of record for content porting (no-wait policy: rebase onto main after the lower PRs merge). **Interaction flags**: F2's #1065 dispose-retention test must be re-verified after the stack merges (new `dispose()` now drains tasks); F3's Task.ts port must be verified against the new main's 23-hunk delta; F1a/F1b spec content has no textual conflict (PRs MERGEABLE). **2026-09-07**: upstream/main has further advanced to `4c7474d42` (re-verified; the interaction flags above remain the watch items). **Rebase (post-#1425)**: upstream/main advanced `4c7474d42` → `a3e31e14b` (2026-09-07); the whole F0–F7 chain was re-stacked onto `a3e31e14b` in one pass and force-pushed (2026-09-07). Only series conflict: `src/eslint-suppressions.json` at F3 — main's #1425 had lowered `core/webview/ClineProvider.ts` no-explicit-any 12→8 while the series had 12→11; resolved to **7** (= main's 8 − F3's net −1; F4–F7 proven neutral via parent-chain arithmetic), the `ClineProvider.sticky-mode.spec.ts` entry auto-merged at 33; both verified with the repo's own `pnpm exec eslint --prune-suppressions --max-warnings=0` gate on the final tree (exit 0, count held, 346 entries semantic-identical to the committed blob) — suppression counts never increased. New heads (the table's State and Stack-base cells carry the re-stacked chain): F0 `e6bc7a294` · F1a `3c43a9e7c` · F1b `a37d0850e` · F1c `6021fee5d` · F2 `92b109654` · F3 `7d91ff4bd` · F4 `da97f98ee` · F5 `cfa678f1c` · F6 `09876aeea` · F7 `ac364cb20`; chain parents verified via `rev-parse <sha>^` (F0′ on main, each next on the previous). The series diff content is byte-identical to the pre-rebase series (#1425 touches no series file), so all local gate evidence (mutation/stryker/lint) remains valid; CI re-runs on the new heads. F0 #1528 marked ready for review 2026-09-07 — the serial bot-wait sequence (F0→F1a→…→F7) is in progress; the stale `has-conflicts` labels on #1557/#1560/#1562/#1565/#1567 (pre-rebase) are cleared only by the repo's reconcile workflow — label removal is admin-only for this account (403 verified), so no manual removal attempted. **Bot-flow calibration (F0, 2026-09-07)**: the shared CI runner queue holds ~110 jobs across the 10 new heads (~3/min drain observed: mutation-diff 06:14, windows unit-test 06:18, e2e-mock 06:22 — 11 F0 jobs still QUEUED at 06:35); the `Label PR review state` labeler self-cascades via its own `workflow_run` events (~25s runs superseding each other, no PR context, no label changes) while its `pull_request`/`issue_comment`-triggered runs queue behind the CI jobs — so F0's labels (incl. the 5 stale `has-conflicts` clears) are expected to land ~30+ min after the last push; CodeRabbit re-review of the new head in progress since 06:11:50Z; F0 = the serial sequence's calibration PR before F1a #1546 is marked ready. **Required-check ruleset (2026-09-07, public `/rules/branches/main`)**: the required status checks are exactly `check-translations`, `platform-unit-test (ubuntu-latest)`, `platform-unit-test (windows-latest)`, `compile`, `knip`, `e2e-mock` — the labeler's `awaiting-maintainer` needs all six green on the head + a fresh CodeRabbit APPROVED review on the head sha + mergeable. Consequences: (a) `mutation-diff` is **advisory** — F4 #1562's deterministic gate failure (509 > 500 changed executable lines; the cumulative F0–F4 stack diff) does NOT block the serial bot pass (visible to the maintainer; the gate's own guidance: "Split the PR or obtain a maintainer-reviewed narrow exclusion"; F5–F7 cumulative diffs exceed the same advisory limit and their mutation runs are expected non-blocking failures). (b) `e2e-mock` IS required, so the 3 red new-head e2e runs on draft heads would have blocked their PRs' passes: F1b #1550 (60s timeouts on `search_files no matches` + `terminal reuse zero-chunk`), F6 #1565 and F7 #1567 (30s timeout on the F6/F7 3-panel coordinated-round test) — all load flakes (F0's identical suite passed 87/87 on the same queue). `gh run rerun` is admin-only (403), so each was re-triggered with an **empty no-op commit** (0-line diff; upper-stack branches unaffected — F1c–F6' remain parented on the pre-empty-commit F1b/F6 heads): F1b `a37d0850e` → `dfc87297f`, F6 `09876aeea` → `3d96e65f1`, F7 `ac364cb20` → `b8b6ef39c` (2026-09-07; all 3 PRs re-verified MERGEABLE on the new heads). **Serial progress (2026-09-07 ~06:45Z)**: F0 #1528 required checks 3/6 green (windows unit-test 06:18, e2e-mock 06:22, knip 06:25); `check-translations` / `platform-unit-test (ubuntu-latest)` / `compile` still queued inside F0's own Code QA run (06:11:47 — ahead of the 06:31 empty-commit runs), CodeRabbit re-review in progress since 06:11:50Z, labels still `[]`. Labeler mechanics confirmed working: 2 SUCCESS reconcile runs (06:12:10 issue_comment, 06:26:13 workflow_run) — the stale `has-conflicts` on **#1567 was cleared by the labeler** (only #1557/#1560/#1562/#1565 remain; they clear when a PR-context or associated run reconciles each branch — no action needed; drain ≈1–2 real runs/min). New-head e2e outcomes so far: F5 #1557 **passed** (cfa678f1c), F4 #1562 in progress, F3 #1560 queued (flake risk — if it flakes, same empty-commit lever applies, severing the F3/F4 parent seam in the same harmless pattern as F1b/F1c and F6/F7). CodeRabbit has started fresh reviews of the new F1b+/F6+/F7+ heads (status comments 06:31–06:32Z). F1a #1546 stays draft until F0 reaches `awaiting-maintainer`. **Serial progress (2026-09-07 ~06:38Z, round 4)**: CodeRabbit mechanism fully decoded from `.github/workflows/label-pr-review-state.yml` (L55–89, L250–307, L608–747) — CodeRabbit auto-review is LABEL-GATED (`.coderabbit.yaml`: required label `coderabbit-review-active`; the 09-04 "Review skipped" comment on F0's old head confirms it). The labeler: (a) while any required check is pending/failed → strips state labels AND removes `coderabbit-review-active` (L610–621) — this is exactly why F0's labels are `[]` right now; (b) when all required checks pass on head + not draft + no fresh CodeRabbit APPROVED → sets `activateCodeRabbit=true` + `recycleCodeRabbitLabel` (L695–696) → adds `coderabbit-review-active` + `awaiting-coderabbit` (L747); (c) CodeRabbit then reviews the head sha and submits a formal review; (d) a later labeler run (triggered by CodeRabbit's summary comment = `issue_comment`) sees `freshCodeRabbitReview.state === 'APPROVED'` on the head sha (L644–663) + mergeable + not draft → **`awaiting-maintainer`** (L704). So the CodeRabbit "in-progress" status context since 06:11:50Z is the label-wait state, not an active review — it starts only after the required-CI-green labeler run. F0 now 3/6 green; the 3 queued required jobs (`check-translations`, `platform-unit-test (ubuntu-latest)`, `compile`) sit inside F0's own Code QA run 34089750771 (its `Build test VSIX` in_progress since 06:35:10Z; F0's Visual Regression run started in_progress after) — F0's runs are draining ahead of the 06:31 empty-commit runs. Stale `has-conflicts`: #1567 and #1557 cleared by the labeler (5→3; #1560/#1562/#1565 remain). New-head re-trigger runs healthy: F1b+ (dfc87297f) e2e + mutation in_progress; F4 (da97f98ee) e2e in_progress; F6+/F7+ e2e queued. Global queue (100-run window): 22 queued / 6 in_progress / 18 success / 9 failure / 45 cancelled (cancelled = old-head superseded by the empty commits). No action available from this account until F0's 6/6 lands (no `gh run rerun`, no label ops); expect F0 `awaiting-maintainer` roughly 30–90 min after the last required job completes (CI green → labeler label → CodeRabbit review ~15–30 min → APPROVED → next labeler run). **Serial progress (2026-09-07 ~06:45Z, round 5)**: found and pre-empted 4 more flaky new-head runs before their serial turns — **F1c #1552 / F2 #1555 / F3 #1560 e2e-mock failed at the `Run mocked E2E tests` step** (load flake, same signature as F1b; F0/F5's identical suites passed in the same 06:11–06:12Z window) and **F5 #1557's `platform-unit-test (windows-latest)` failed at the `Save Turbo cache` step** (the test step itself passed — infra flake; all other Code QA jobs green). Re-triggered all 4 with empty no-op commits (0-line diff; each pre-push `check-types` 11/11 passed): F1c `6021fee5d`→`9712eb25d`, F2 `92b109654`→`9b7c701fe`, F3 `7d91ff4bd`→`da47c6977`, F5 `cfa678f1c`→`5d0a09b89` — all 4 re-verified **MERGEABLE** via `gh pr view`. Pushes hit a burst of github.com:443 connection flakes (1–3 retries each, all eventually exit 0; F1c pushed from `vps2-f1c-wt` since that branch is checked out there, F2/F3/F5 from `vps2-rebase`). The mutation-gate failures on new heads (F3/F4/F5/F6+/F7+ cumulative diffs > 500 changed lines) confirmed as the known ADVISORY fast-fails — non-blocking. F0 #1528 at round 5: 4/6 required green, `compile` + `platform-unit-test (ubuntu-latest)` in_progress since 06:37:53Z inside its own Code QA run — the 2 jobs on F0's critical path to `awaiting-maintainer`. All 10 heads now: F0 `e6bc7a294` / F1a `3c43a9e7c` / F1b `dfc87297f` / F1c `9712eb25d` / F2 `9b7c701fe` / F3 `da47c6977` / F4 `da97f98ee` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c` (all MERGEABLE). **Serial progress (2026-09-07 ~06:46Z, round 6)**: F0 #1528 **6/6 required checks green** (all 6 COMPLETED SUCCESS: check-translations 06:37, ubuntu unit-test 06:37, windows unit-test 06:18, compile 06:37, knip 06:25, e2e-mock 06:22). Labels still `[]`, reviews `[]`, CodeRabbit context still `in_progress` (label-wait since 06:11:50). The 6th required check (ubuntu unit-test) went green ~06:44–06:45, which should have triggered a `workflow_run` labeler event; indeed a labeler `workflow_run` run was created at 06:46:43Z (pending) but is part of a self-cascade where 11 labeler runs in the 06:42–06:46 window were cancelled by newer events (the 4 empty-commit pushes + their CI completions). 2 labeler runs remain: `pending [workflow_run] 06:46:43Z id=34092347288` (most likely to complete and act on F0) and `queued [pull_request_target] 06:41:42Z id=34091974282` (F1c's push event, stuck in runner queue ~5 min). When either completes, the labeler should see F0 6/6 + not draft + no CodeRabbit review → add `coderabbit-review-active` + `awaiting-coderabbit` → CodeRabbit starts reviewing head `e6bc7a294`. 4 re-trigger heads CI status: F1c (9712eb25d) e2e in_progress + mutation in_progress + Code QA queued; F2 (9b7c701fe) e2e in_progress + mutation in_progress + Code QA queued; F3 (da47c6977) e2e in_progress + Code QA queued + mutation queued; F5 (5d0a09b89) CodeQL in_progress, all else queued. No new flakes observed. F1a #1546 stays draft until F0 reaches `awaiting-maintainer`. **Serial progress (2026-09-07 ~06:56Z, round 7)**: diagnosed why F0 (6/6 on `e6bc7a294`) never got bot labels: the labeler's PR-selection logic (decoded from `label-pr-review-state.yml` L100–149) reconciles ONLY the event-triggering PR(s) for `workflow_run`/`pull_request_target`/`issue_comment` runs (eventPrNumbers); full-repo reconciliation happens only for main-branch pushes, the schedule trigger, and workflow_dispatch. F0's own CI-completion events were all consumed by cascade-cancelled runs: E2E completion 06:23:09Z → run 34090591922 cancelled 1s later; Code QA completion 06:46:30Z → runs 34092334369/34092342380/34092347288 all cancelled. The live base-repo workflow DOES have a schedule trigger (last success 05:22:15Z; daily pattern, not hourly — our worktree copy lacks it) and a push trigger (02:36:15Z success), so without action F0 would have waited until the next daily schedule (~05:22Z) or a maintainer. Levers checked and ruled out: `workflow_dispatch` via API = 403 (admin rights required); all 5 CODEOWNERS (`taltas` `JamesRobert20` `navedmerchant` `hannesrudolph` `edelauna`) already review-requested on F0 so no fresh `review_requested` event is possible; no org teams exist to request. **Lever used**: empty no-op commit on F0 (06:55Z, pre-push `check-types` 11/11 cached) → new head `591805850` (parent `e6bc7a294`, 0-line diff) → fired a `synchronize` pull_request_target labeler run (06:55:30Z, in_progress at snapshot) and re-triggered F0's full CI on the new head. F0 6/6 on `591805850` expected ~07:15–07:30Z; the last trigger-workflow completion (Code QA) then fires an F0-targeted labeler run that should see 6/6 + not-draft + no fresh CodeRabbit APPROVED → add `coderabbit-review-active` + `awaiting-coderabbit` → CodeRabbit reviews head `591805850` (~15–30 min) → `awaiting-maintainer`. F5 #1557's new-head `Changed-code mutation testing` failed at 06:44:37Z (advisory gate, cumulative diff > 500 changed lines — the known non-blocking fast-fail). 4 re-trigger heads at 06:49Z: F1c/F2/F3/F5 e2e all in_progress, no new blocking flakes. All 10 PRs re-verified MERGEABLE post-push: F0 `591805850` (isDraft=false) / F1a `3c43a9e7c` / F1b `dfc87297f` / F1c `9712eb25d` / F2 `9b7c701fe` / F3 `da47c6977` / F4 `da97f98ee` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. **Serial progress (2026-09-07 ~07:01Z, round 8)**: two more e2e-mock load flakes confirmed and re-triggered. F1c #1552's e2e on head `9712eb25d` (run 34091976148) and F2 #1555's e2e on head `9b7c701fe` (run 34092036909) both failed at the `Run mocked E2E tests` step with an identical signature: 83 passing / 2 failing, the two failures being `Roo Code search_files Tool → Should handle search with no matches` and `Terminal reuse after zero-chunk shell race → completes two sequential zero-chunk commands on a reused terminal`, each `Error: Timeout after 60s` at `out/suite/utils.js:15:20` — both tests unrelated to the F1c/F2 diff (webview identity / durable viewstates), same load class as the documented 60s timeouts (pre-existing; the timing out test varies per run). Re-triggered both with empty no-op commits (pre-push `check-types` 11/11 cached): F1c `9712eb25d`→`40f538f2b` (pushed from `vps2-f1c-wt`; first push attempt exit 128 github.com:443 flake, retry exit 0), F2 `9b7c701fe`→`3dcf4bfe3` (from `vps2-rebase`; first ls-remote attempt flaked, retry ok). Both synchronize labeler runs completed SUCCESS; new-head CI re-triggered (F1c e2e in_progress 07:00Z, F2 e2e queued 07:00:30Z). F5 #1557's e2e on head `5d0a09b89` is now **success** (round-5 re-trigger fixed the windows unit-test flake and e2e passed; its only remaining failure = the advisory mutation gate). F3 #1560's e2e on `da47c6977` still in_progress (started 06:43:21Z) — verdict pending. F0 #1528 on new head `591805850`: 4/6 required green at snapshot (check-translations / knip / compile / ubuntu unit-test all SUCCESS from the 06:55Z start — turbo cache had ubuntu unit-test finish in <1 min); `e2e-mock` + `platform-unit-test (windows-latest)` in_progress, 6/6 expected ~07:10–07:25Z. All 10 PRs MERGEABLE. Head list: F0 `591805850` / F1a `3c43a9e7c` / F1b `dfc87297f` / F1c `40f538f2b` / F2 `3dcf4bfe3` / F3 `da47c6977` / F4 `da97f98ee` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. **Serial progress (2026-09-07 ~07:07Z, round 9)**: F1a #1546's CI is now **6/6 required green** (all SUCCESS on head `3c43a9e7c`; mutation gate also green) — the first PR in the chain with fully green CI. Two more e2e-mock load flakes found and re-triggered. F1b #1550's e2e on re-trigger head `dfc87297f` (run 34091203542, failed 06:40:05Z) — same load-class signature: 83 passing / 2 failing, 60s timeouts — re-triggered with an empty commit `dfc87297f`→`d6789c0b8` (from `vps2-f1b-wt`; new-head CI in_progress 07:05:27Z). F4 #1562's e2e on `da97f98ee` (run 34089757832, failed 06:44:16Z after a 33-minute run that started 06:11Z during peak concurrent e2e load) — heavier load-class signature: 72 passing / 13 failing, suite time 14m (vs ~5m normal), all failures being provider e2e task-completion tests (`Bedrock provider … Claude Opus 4.8`, `DeepSeek V4 flash/pro … tool-using task with reasoning enabled`, `Gemini provider gemini-3-flash-preview … reasoning effort`, `xAI provider … xAI Responses API` …) each timing out at 30s/180s — pure runner-overload timeouts unrelated to the F4 diff; F0 and F5 passed the same suites in the same window. Re-triggered F4 with an empty commit `da97f98ee`→`b6aea7e8d` (from `vps2-rebase`; new-head CI queued 07:06:07Z). F0 #1528 on `591805850`: 5/6 required green at snapshot (e2e-mock turned SUCCESS ~07:02Z; only `platform-unit-test (windows-latest)` in_progress, expected green ~07:05–07:15Z). F3 #1560's e2e on `da47c6977` still in_progress (started 06:43:21Z). F5 #1557's e2e on `5d0a09b89` is green (only the advisory mutation failure remains). All 10 PRs MERGEABLE. Head list: F0 `591805850` / F1a `3c43a9e7c` / F1b `d6789c0b8` / F1c `40f538f2b` / F2 `3dcf4bfe3` / F3 `da47c6977` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. **Serial progress (2026-09-07 ~07:10Z, round 10)**: F1c #1552's e2e failed a second time on re-trigger head `40f538f2b` (run 34093336590, failed 07:07:07Z; 7-min run) with the identical load-class signature — 83 passing / 2 failing, the same two tests (`Roo Code search_files Tool → Should handle search with no matches` and `Terminal reuse after zero-chunk shell race → … reused terminal`, both `Error: Timeout after 60s`), plus `[Terminal Process] Stream error: VSCE shell integration stream did not start within 5 seconds`. F3 #1560's e2e also failed on `da47c6977` (run 34092094608, failed 07:04:19Z; 21-min run) with the same heavy load-class profile as F4: 72 passing / 13 failing, suite 15m (vs ~5m normal), provider e2e task-completion timeouts at 30s/180s. Load-window correlation: every e2e failure so far belongs to a run started 06:11–06:44Z (peak concurrency, 8+ e2e suites); runs started 06:44Z+ (F5 `5d0a09b89`, F0 `591805850`) passed the full suite in the same window — consistent with runner-overload flakes, not a diff regression (F1c's diff is webview identity, F3's is mode/profile writes; neither touches search_files or terminal fixtures). Re-triggered both with empty no-op commits (pre-push hooks passed): F1c `40f538f2b`→`1c90a81b2` (3rd re-trigger, from `vps2-f1c-wt`, pushed 07:08Z), F3 `da47c6977`→`f930a24ad` (1st re-trigger, from `vps2-rebase`, pushed 07:09Z); new-head CI queued 07:08:27Z / 07:09:49Z; both synchronize labeler runs SUCCESS. F0 #1528 on `591805850`: 5/6 required green at snapshot — `platform-unit-test (windows-latest)` and the `Code QA Roo Code` workflow (started 06:55:43Z) still in_progress; 6/6 expected ~07:15–07:30Z, after which F0's own trigger-workflow completion fires the F0-targeted labeler run for `coderabbit-review-active` + `awaiting-coderabbit`. E2E suites still running at snapshot: F1b `d6789c0b8`, F1c `1c90a81b2`, F2 `3dcf4bfe3`, F3 `f930a24ad`, F4 `b6aea7e8d`. All 10 PRs MERGEABLE. Head list: F0 `591805850` / F1a `3c43a9e7c` / F1b `d6789c0b8` / F1c `1c90a81b2` / F2 `3dcf4bfe3` / F3 `f930a24ad` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. **Serial progress (2026-09-07 ~07:12Z, round 11)**: F2 #1555's e2e failed a second time on re-trigger head `3dcf4bfe3` (run 34093379072, failed 07:08:06Z; 7.5-min run) with the same load-class signature: 83 passing / 2 failing, the same two tests (`Roo Code search_files Tool → Should handle search with no matches`, `Terminal reuse after zero-chunk shell race → … reused terminal`), both `Error: Timeout after 60s` — F2's diff (durable viewStates pruning) does not touch search/terminal. Re-triggered with an empty commit `3dcf4bfe3`→`8dd2a17ae` (from `vps2-rebase`, pushed 07:11Z; new-head CI queued 07:11:45Z; synchronize labeler run SUCCESS). Load-window correlation refined: failures so far belong to runs started 06:11–06:44Z (peak concurrency) and to two runs started 07:00Z (F1c #2 `40f538f2b`, F2 #2 `3dcf4bfe3`); runs started 06:44–06:55Z (F5 `5d0a09b89`, F0 `591805850`) passed the full suite in the same window. Every failure shows the same two borderline 60s tests timing out — consistent with runner contention, not a diff regression (no diff in the series touches search/terminal fixtures). F4 #1562's new-head `Changed-code mutation testing` failed (advisory gate, cumulative >500-line fast-fail — known non-blocking). F0 #1528 on `591805850`: 5/6 required green at snapshot — `platform-unit-test (windows-latest)` in_progress since 06:55:34Z and the `Code QA Roo Code` workflow still in_progress since 06:55:43Z; 6/6 expected ~07:15–07:30Z. E2E suites running at snapshot: F1b `d6789c0b8`, F1c `1c90a81b2`, F3 `f930a24ad`, F4 `b6aea7e8d` + F2 `8dd2a17ae` (queued). All 10 PRs MERGEABLE. Head list: F0 `591805850` / F1a `3c43a9e7c` / F1b `d6789c0b8` / F1c `1c90a81b2` / F2 `8dd2a17ae` / F3 `f930a24ad` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. **Serial progress (2026-09-07 ~07:18Z, round 12)**: F1b #1550's e2e failed a second time on re-trigger head `d6789c0b8` (run 34093770028, failed 07:12:25Z; 7-min run) with the identical load-class signature: 83 passing / 2 failing, the same two tests (`Roo Code search_files Tool → Should handle search with no matches`, `Terminal reuse after zero-chunk shell race → … reused terminal`), both `Error: Timeout after 60s` — F1b's diff (getState merge) does not touch search/terminal. Re-triggered with an empty commit `d6789c0b8`→`019c84dcf` (3rd re-trigger, from `vps2-f1b-wt`, pushed 07:17Z; new-head CI started 07:17:22Z; synchronize labeler run SUCCESS). **F0 #1528 label progress**: F0's own `Code QA Roo Code` run completed SUCCESS 07:12:14Z; the F0-targeted labeler run it spawned (workflow_run event, queued since 07:12:17Z, held under runner contention while e2e suites ran, completed 07:17:08Z) applied labels `coderabbit-review-active` + `awaiting-coderabbit` to F0. F0 state now: 6/6 required green + mutation SUCCESS + not draft + MERGEABLE + labeled. Next step: CodeRabbit auto-review (label-gated) reviews head `591805850` — the fresh review must carry `commit_id === 591805850…`; once its state reaches APPROVED the next labeler run applies `awaiting-maintainer` (terminal bot-pass state), after which serial policy runs `gh pr ready 1546` (F1a). E2E suites running at snapshot: F1b `019c84dcf` (just started), F1c `1c90a81b2`, F3 `f930a24ad`, F4 `b6aea7e8d` + F2 `8dd2a17ae` (queued). All 10 PRs MERGEABLE. Head list: F0 `591805850` / F1a `3c43a9e7c` / F1b `019c84dcf` / F1c `1c90a81b2` / F2 `8dd2a17ae` / F3 `f930a24ad` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. **Serial progress (2026-09-07 ~07:27Z, round 13)**: F1c #1552's e2e failed a third time on `1c90a81b2` (run 34094008573, failed 07:17:06Z; 9-min run) with the identical load-class signature — 83 passing / 2 failing, same two tests (`search_files Tool → no matches`, `Terminal reuse after zero-chunk shell race`), both `Error: Timeout after 60s`. F2 #1555's e2e failed a third time on `8dd2a17ae` (run 34094281444, failed 07:19:26Z; 7.5-min run) with the identical 83/2 signature. F4 #1562's e2e failed a second time on `b6aea7e8d` (run 34093822847, failed 07:23:14Z; 17-min run) with the heavy overload-class signature — 13 failing, provider task-completion timeouts at 30s/180s, same as its first failure. Load-class taxonomy now: (a) 83/2 with 2×60s timeouts on the two borderline tests under 3–4 concurrent e2e suites (F1c×3, F2×3, F1b×2); (b) 72/13 with 30s/180s provider timeouts on 14–17-min runs (F3 first head, F4×2). Runs under ~3 concurrency passed (F5, F0, F1a). **Staggered re-trigger strategy**: pushed F1c's 4th re-trigger `1c90a81b2`→`03c46f714` (from `vps2-f1c-wt`, 07:25:59Z; new-head CI started immediately, e2e in_progress) into the slot freed by F4's completion; F2's 4th and F4's 2nd re-triggers deliberately deferred to keep concurrent e2e suites ≤3 (F1c is earlier in chain order than F2/F4). F0 #1528: `coderabbit-review-active` + `awaiting-coderabbit` labels applied 07:17:08Z; CodeRabbit review of head `591805850` not yet posted at 07:25Z (label applied 07:17:08Z — CodeRabbit typically 15–30 min → expected ~07:32–07:47Z); 6/6 required green. E2E in flight at snapshot: F1b `019c84dcf` (started 07:17:22Z), F3 `f930a24ad` (started 07:09:49Z, 14-min overload class), F1c `03c46f714` (started 07:25:59Z). All 10 PRs MERGEABLE. Head list: F0 `591805850` / F1a `3c43a9e7c` / F1b `019c84dcf` / F1c `03c46f714` / F2 `8dd2a17ae` / F3 `f930a24ad` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. **Serial progress (2026-09-07 ~07:50Z, round 14)**: F0 #1528: CodeRabbit posted a CHANGES_REQUESTED review of head `591805850` (2026-09-07T07:26:51Z, run `4afe1273-8739-4235-90d3-311db5f6ccb9`, ASSERTIVE, 2 actionable comments) and the labeler flipped F0 to `awaiting-author`. Both comments verified valid against the code: (a) `openClineInNewTab` race — two concurrent calls can each find no tab panel and create two; (b) test hygiene — `@ts-ignore` private `view` assignment in `ClineProvider.spec.ts` and `as any` fixture attachment in `registerCommands.spec.ts`. Fix implemented in `vps2-f0-wt` (4 files, +123/−88): `registerCommands.ts` tracks in-flight tab-panel creation in a module-level `pendingTabPanelCreation` promise (early-return when present; `try/finally` clears once settled) + new `Promise.all` concurrent-open regression test (one panel, one provider); `ClineProvider.spec.ts` sets the view via the public `resolveWebviewView()` instead of `@ts-ignore`; `registerCommands.spec.ts` types `evictCurrentTask`/`refreshWorkspace` as `Mock` on the fixture (drops `as any`); `eslint-suppressions.json` prunes the `registerCommands.spec.ts` entry (count 2→0; file kept in tab format — the `--prune-suppressions` plugin's full-file 3457-line reformat was rejected as out-of-scope). Gates: eslint `--prune-suppressions --max-warnings=0` EXIT:0 on all three touched TS files; vitest 195/195 on the two touched suites. Committed `8d0d44bc3`, pushed 07:47Z (first push attempt hit a transient github.com:443 connect failure; retry OK). PR head now `8d0d44bc3` (MERGEABLE, not draft, labels cleared); full CI + fresh CodeRabbit review will run on the new head — the prior CHANGES_REQUESTED on `591805850` is moot. F1b #1550: 3rd e2e failure confirmed the 83/2 load class (run 34094742771 on `019c84dcf`, failed 07:25:41Z, ~8 min, same two tests, 60s timeouts); pushed 4th re-trigger `6d13672b4` (0-line). F3 #1560: e2e failed (run 34094119260, failed 07:27:23Z, ~17.5 min; 72 passing/13 failing, 30s/180s provider timeouts — overload class) AND `platform-unit-test (windows-latest)` failed in the run named `Code QA Roo Code` (run 34094119188, failed 07:25:12Z) — all three coverage steps succeeded; the failure was the `Save Turbo cache` step (actions/cache infra flake, not a test failure); advisory mutation failure (34094119123) noted. Pushed F3 re-trigger `78f2a8296` (0-line; re-runs e2e + windows unit test). F1c #1552: 4th e2e failure (run 34095436919, failed 07:33:09Z) — 83/2 load class again; 5th re-trigger deferred under the stagger rule. E2E pool after pushes (≤3 concurrent): F0 `8d0d44bc3` (starting), F1b `6d13672b4` (queued), F3 `78f2a8296` (queued); deferred: F1c 5th, F2 4th, F4 2nd — next pushes only when pool ≤2. Post-push sweep 07:50Z: all 10 PRs MERGEABLE. Head list: F0 `8d0d44bc3` / F1a `3c43a9e7c` / F1b `6d13672b4` / F1c `03c46f714` / F2 `8dd2a17ae` / F3 `78f2a8296` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`. Round 15 (2026-09-07 ~08:04Z): F0 #1528 @ `8d0d44bc3` — all 6 required checks now green (e2e-mock run 34097208497 07:52:56Z; platform-unit-test ubuntu+windows; compile; knip; check-translations); `mutation-diff` FAILURE (run 34097208546) verified **advisory-only** (branch ruleset `/rules/branches/main` re-queried: `required_status_checks` = exactly the 6 contexts above; mutation-diff not required ⇒ does not block bot pass) — cause: the `8d0d44bc3` IIFE wrapper re-indented the whole `openClineInNewTab` creation body, widening the diff-based changed-lines scope to ~38 previously-out-of-scope body lines (23 survived + 15 uncovered; the un-pinned lines: MdmService catch branch, unused ternary branches, event-listener callbacks, un-asserted strings/commands — pre-existing upstream lines the tests never pinned, surfaced only by the re-indent); per series policy no content change for advisory gates. Labels now `awaiting-coderabbit` + `coderabbit-review-active` (labeler re-queued after push) → fresh CodeRabbit review in flight (ETA 10–30 min); `awaiting-maintainer` expected on fresh APPROVED. F1b #1550: 4th e2e failure = 83/2 load class again (run 34097300971, failed 07:55:42Z, ~7 min, `83 passing` + the same two 60s-timeout tests) → pushed 5th re-trigger `e75806c44` (0-line; first push attempt hit a github.com:443 connect flake, then two pwsh arg-parsing errors on the refspec/lease args — retry with variable-built args OK). F1c #1552: pushed 5th re-trigger `acb18b2f9` (0-line; pool was 1 after F1b's 4th failure — staggered, chain order F1b→F1c). **Plan adjustment (user asked 08:1xZ)**: the stagger threshold is tightened from pool ≤3 to **max 2 concurrent e2e suites** (push a re-trigger only when pool ≤1) — the 07:47–07:55Z window at pool 3 produced both outcomes simultaneously (F0 e2e green run 34097208497 / F1b red run 34097300971), so 3 carries no safety margin. Re-trigger pushes are also serialized (one push per e2e verdict, not batched — the F1b+F1c same-window push above was the last batch). No content/test changes: the 83/2 class is infra load-flake (60s timeouts on the same two tests across four different PR heads; F1a's 999-line ClineProvider diff and F5 e2e-passed at low load ⇒ series content does not regress those tests) — the two e2e tests and all series diffs stay untouched. Cost: ≈+25 min/PR in the flake loop for the 5 PRs still needing e2e green (F1b/F1c/F2/F3/F4); F6/F7 e2e verdicts still to be verified. F3 #1560: 5th e2e in progress (run 34097307741, started 07:48:36Z; >17 min elapsed = overload-class risk). E2E pool now 3/3 (F1b 5th, F1c 5th, F3 5th); deferred: F2 4th, F4 2nd (push only when pool ≤2). Post-push sweep 08:04Z: all 10 PRs MERGEABLE (#1550 head `e75806c44`, #1552 head `acb18b2f9`). Head list: F0 `8d0d44bc3` / F1a `3c43a9e7c` / F1b `e75806c44` / F1c `acb18b2f9` / F2 `8dd2a17ae` / F3 `78f2a8296` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`.
Round 16 (2026-09-07 ~08:20Z): F0 #1528 — CodeRabbit posted CHANGES_REQUESTED on `8d0d44bc3` at 08:06:48Z (run e2bbcee2-0b0f-4a64-9d67-013016a46f97; review scope `591805850`→`8d0d44bc3`, 4 files), 2 actionable comments: (1) strengthen the serialization test so both results are pinned against the constructed ClineProvider instance exposed by the mocked constructor (identity + not-undefined); (2) add focused tests for the creation branches — MdmService.getInstance throw→undefined fallback, empty visibleTextEditors→newGroupRight+ViewColumn.Two, undefined viewColumn→zero fallback, `new ClineProvider(..., "editor", ...)` argument — or correct the validation claim. Implemented all (satisfies CodeRabbit AND kills the 38 advisory mutation-diff mutants, all on the same registerCommands.ts L322–383 body lines). Commit `c51846007` `test(activate): pin openClineInNewTab creation branches and strengthen the concurrency assertion` (1 file, +118/−4): 5 new tests in the `openClineInNewTab` describe (MdmService-throw fallback, empty-editors newGroupRight+ViewColumn.Two+iconPath+lockEditorGroup, undefined-viewColumn zero-fallback→column 1, "editor"-ctor-arg with sentinel MdmService, visibility/dispose listener wiring) + serialization test now pins both results against `ClineProvider.mock.instances[0]`. Local gates: eslint EXIT:0 (no suppression change), tsc EXIT:0, vitest 200/200 (was 195). First push blocked by pre-push check-types (TS2352: readonly `window.visibleTextEditors` cast rejected) → `as unknown as` fix + amended. Pushed 08:15Z; sweep: all 10 PRs MERGEABLE, #1528 not draft, labels `[]` (labeler re-queued 08:15:38Z); new CI in flight @ `c51846007` (e2e-mock 34099653035, mutation 34099653089, Code QA, CodeQL, visual, release-validation). E2E pool verdicts 08:15Z: F1b 5th `34098482108` FAILURE (83/2 class: `83 passing` + same two 60s-timeout tests, ~7.5 min), F1c 5th `34098494613` FAILURE (83/2 class, ~7.3 min), F3 5th `34097307741` FAILURE (overload class: 13 failing, 30s/180s timeouts, ~19 min). F6/F7 e2e failure signatures confirmed 87/1 class (`Timed out after 1 coordinated rounds`, 30s, ~4 min): F6 `34091269918`, F7 `34091233446` — queued behind pool rule. Pushed F1b 6th no-op re-trigger `e99726fbd` (08:17Z; pool was 1 = F0 e2e only; ready-chain order F1b→F1c). Post-push sweep: all 10 MERGEABLE (#1550 head `e99726fbd`). Re-trigger queue (serialized, one push per verdict, ready-chain order): F1c 6th → F2 4th → F3 6th → F4 3rd → F6 → F7. Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer`. Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `e99726fbd` / F1c `acb18b2f9` / F2 `8dd2a17ae` / F3 `78f2a8296` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`.
Round 17 (2026-09-07 ~08:26Z): F0 #1528 @ `c51846007` — required 5/6 green: e2e-mock SUCCESS on the first run at the new head (no load flake), check-translations/compile/knip/platform-unit-test (ubuntu) green; platform-unit-test (windows) in progress. `mutation-diff` advisory FAILURE, surviving mutants 38→7 (full locations from the `changed-code-mutation-report` artifact): L324–327 MdmService catch block→{}, L330 `Math.max`→`Math.min` (viewColumn spread), L330 `viewColumn || 0`, L340 `Math.max(lastCol+1,1)`→`Math.min`, L340 `lastCol+1`, L349/L375 `"tab"` surface tag→"". 3 of 7 (the catch block — `let mdmService` is uninitialized so catch{} ≡ catch{mdmService=undefined}; the two `"tab"` tags — `setPanel` branches on `type === "sidebar"` so `""` ≡ `"tab"`) are behaviorally equivalent and unkillable from the spec without production changes; the other 4 are killable but the gate stays red either way (≥3 survivors), and `mutation-diff` is not in the 6 required checks gating `awaiting-maintainer` (it has been advisory-RED at all three F0 heads: `591805850`/`8d0d44bc3`/`c51846007`) → per series policy NO content change for advisory gates, no push. F1a #1546 @ `3c43a9e7c`: ALL checks green including `mutation-diff` SUCCESS. CodeRabbit: no fresh review posted on `c51846007` yet (review history: CHANGES_REQUESTED ×2 on `591805850` 07:26Z / `8d0d44bc3` 08:06Z) — in flight (labels `[]`, `coderabbit-review-active` pending), ETA 10–30 min after that label lands. E2E pool: F1b 6th in progress (`e99726fbd`, e2e started 08:18Z) → pool 1; pushed F1c 6th no-op re-trigger `18122c3c1` (08:25Z, pool 1→2 under the adjusted stagger rule: max 2 concurrent, push only when pool ≤1). Re-trigger queue (serialized, one push per verdict, ready-chain order): F2 4th → F3 6th → F4 3rd → F6 → F7. Post-push sweep 08:26Z: all 10 PRs MERGEABLE (#1552 head `18122c3c1`). Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer` (then `gh pr ready 1546` → chain). F7 `b8b6ef39c`.
Round 18 (2026-09-07 ~08:30Z): F0 #1528 @ `c51846007` — still 5/6 required (platform-unit-test windows in progress since 08:15Z window); labels still `[]` (no fresh CodeRabbit review posted on `c51846007` yet — in flight). F1b #1550 6th e2e FAILURE (run 34099886836, 08:25Z, ~7.4 min) = 83/2 class again (`83 passing` + same two 60s-timeout tests) — 6 confirmed load-flake failures; `mutation-diff` SUCCESS on `e99726fbd`. Pool was 1 (F1c 6th e2e in progress since ~08:25Z) → pushed F2 4th no-op re-trigger `f4621e82e` (08:29Z, ready-chain order F2 next; e2e started ~08:29Z, pool now 2). F1c #1552 6th e2e still in progress on `18122c3c1`. Re-trigger queue (serialized, one push per verdict): F3 6th → F4 3rd → F6 → F7. Post-push sweep 08:30Z: all 10 PRs MERGEABLE (#1555 head `f4621e82e`). Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer` (then `gh pr ready 1546` → chain; F1a #1546 is fully green incl. mutation-diff and ready the moment the chain reaches it). Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `e99726fbd` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `78f2a8296` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`.
Round 19 (2026-09-07 ~08:30Z): F0 #1528 @ `c51846007` — ALL 6 required checks now green (platform-unit-test windows completed success ~08:28Z; e2e-mock/check-translations/compile/knip/unit-ubuntu already green) and the reconcile workflow re-queued the bot: labels now `awaiting-coderabbit` + `coderabbit-review-active` (08:30Z). Fresh CodeRabbit review on `c51846007` expected within 10–30 min of that label (prior two reviews were CHANGES_REQUESTED on `591805850` 07:26Z and `8d0d44bc3` 08:06Z — both actionable items fixed in `c51846007`); APPROVED → `awaiting-maintainer` = serial chain step 1 unblocked (`gh pr ready 1546`). Advisory `mutation-diff` stays RED on `c51846007` (7 survivors, 3 behaviorally equivalent — non-blocking per series policy). E2E pool = 2 (max): F1c 6th `18122c3c1` e2e in progress since ~08:25Z, F2 4th `f4621e82e` e2e in progress since ~08:28Z — no new pushes until a verdict lands. Note: F1b's 6th failure (round 18) re-enters the re-trigger queue after F2's verdict (ready-chain order: F1b 7th → F3 6th → F4 3rd → F6 → F7). Serial ready-marking still blocked at step 1 (awaiting F0 `awaiting-maintainer`). No head changes this round.
Round 20 (2026-09-07 ~08:34Z): F1c #1552 6th e2e FAILURE (run 34100554814, 08:26:01Z→08:32:58Z ~6.9 min) = 83/2 class again (`83 passing` + same two 60s-timeout tests) — 6 confirmed load-flake failures; platform-unit-test ubuntu+windows green on `18122c3c1`, `mutation-diff` in progress. Pool dropped to 1 (F2 4th `f4621e82e` e2e still in progress since 08:27:48Z) → pushed F1b 7th no-op re-trigger `8fc4f7609` (08:34Z, ready-chain order: F1b re-enters queue after its 6th failure and precedes F3). Post-push sweep: all 10 PRs MERGEABLE (#1550 head `8fc4f7609`). F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active`, fresh CodeRabbit review still in flight (no 3rd review posted on `c51846007` yet; prior two were CHANGES_REQUESTED). Re-trigger queue (serialized, one push per verdict, ready-chain order): F3 6th → F4 3rd → F6 → F7 (F1b 7th now in flight). Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer`. Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `8fc4f7609` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `78f2a8296` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`.
Round 21 (2026-09-07 ~08:37Z): F2 #1555 4th e2e FAILURE (run 34100711329 on `f4621e82e`, 08:27:48Z→08:34:46Z ~7 min) = 83/2 class (`83 passing` + same two 60s-timeout pair) — 4 confirmed load-flake failures (72/13 overload class seen on earlier runs). Pool → 1 (only F1b 7th `8fc4f7609` e2e in progress) → pushed F3 6th no-op re-trigger `39423f27e` (08:37Z, ready-chain order). Post-push sweep: all 10 PRs MERGEABLE (#1560 head `39423f27e`). F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active` (re-queued ~08:30Z), fresh CodeRabbit review still in flight (no 3rd review posted on `c51846007` yet; prior two CHANGES_REQUESTED). Re-trigger queue (serialized, one push per verdict, ready-chain order): F4 3rd → F6 → F7 (F1b 7th + F3 6th now in flight). Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer`. Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `8fc4f7609` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `39423f27e` / F4 `b6aea7e8d` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`.
Round 22 (2026-09-07 ~08:41Z): F1b #1550 7th e2e FAILURE (run 34101217142 on `8fc4f7609`, job 08:33:32Z→08:40:36Z ~7.1 min) = 83/2 class again (`83 passing` + same two 60s-timeout pair) — 7 confirmed load-flake failures; ubuntu+windows platform-unit-test green, mutation-diff SUCCESS on `8fc4f7609`. Pool → 1 (only F3 6th `39423f27e` e2e in progress) → pushed F4 3rd no-op re-trigger `916900f6a` (08:41Z, ready-chain order). Post-push sweep: all 10 PRs MERGEABLE (#1562 head `916900f6a`). F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active` (re-queued ~08:30Z), fresh CodeRabbit review still in flight (no 3rd review posted on `c51846007` yet; prior two CHANGES_REQUESTED). Re-trigger queue (serialized, one push per verdict, ready-chain order): F6 → F7 (F3 6th + F4 3rd now in flight). Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer`. Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `8fc4f7609` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `39423f27e` / F4 `916900f6a` / F5 `5d0a09b89` / F6 `3d96e65f1` / F7 `b8b6ef39c`.
Round 23 (2026-09-07 ~08:56Z): user asked to rebase the other PRs onto the latest upstream main ("something need to fix also update"). Verified: upstream main = `a3e31e14b` (16 commits 2026-09-04 01:24Z → 09-07 02:25Z incl. #1425, #1349, #1319, #1470, #1525 mutation-infra fix, #1527, #1534 node24, #1488/#1506/#1508 providers, #1509/#1510 labeler; repo last push 03:32Z) and the ENTIRE F0–F7 chain was already re-stacked onto exactly `a3e31e14b` in the earlier single-pass rebase (Main-drift bullet) — `git merge-base a3e31e14b <tip>` = `a3e31e14b` for all 10 branch tips (every head contains the latest main), main unchanged since (re-checked ~08:52Z). No rebase action needed; no content fixes needed (#1425 touches no series file — series diff byte-identical, prior local gates valid). 10/10 MERGEABLE re-verified ~08:52Z (F0 not-draft; F1a–F7 draft pending the serial chain). Stale `has-conflicts` labels: #1560/#1562/#1565 all label-empty at 08:45Z — no stale labels remain on any of the 10. New observation: `mutation-diff` advisory FAILURE on F3 `39423f27e` and F4 `916900f6a` — advisory only (not among the 6 required gates), same 0-line content as the earlier green mutation runs on those branches ⇒ non-deterministic infra, consistent with the known cumulative-diff advisory red; does not block `awaiting-maintainer`. E2E pool 2/2: F3 6th (`39423f27e`, run 34101509856, job started 08:36:47Z) ~19 min elapsed — at the upper bound of the 72/13 overload class (13–19 min), verdict imminent; F4 3rd (`916900f6a`, run 34101891209, 08:41:00Z) in progress. F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active` (08:30Z re-queue), fresh CodeRabbit review still not posted at 08:55Z (>25 min past the 08:30Z re-queue — slower than the 10–30 min norm; no lever available from this account, waiting). Re-trigger queue (serialized, one push per verdict): F6 → F7 (F3 6th + F4 3rd in flight). No head changes this round.
Round 24 (2026-09-07 ~08:57Z): F3 #1560 6th e2e FAILURE (run 34101509856 on `39423f27e`, 08:36:47Z→08:53:43Z ~17 min) = **72/13 overload class** — `Error: 13 tests failed.` with 7 provider timeouts (30s/180s) + 6 probe-tag assertions (Bedrock user-agent header, 2× thinking=disabled probe, probe-tag capture, xAI fast non-reasoning, xAI fast reasoning model ID) — matches F3's prior 72/13 overload run 34097307741; ubuntu+windows unit green on `39423f27e`, mutation-diff advisory failure (non-gate, 0-line content). Content-exonerated: F5 head `5d0a09b89` (superset of F3 content, same new main) e2e GREEN and F0 head `c51846007` (same new main) e2e green ⇒ series content is not the cause; load flake. 6 confirmed F3 failures. Pool → 1 (only F4 3rd in flight) → pushed F6 1st no-op re-trigger `3d96e65f1`→`f8ea2f1a2` (08:57Z, ready-chain order; husky check-types all cache hits). Post-push sweep: all 10 PRs MERGEABLE (#1565 head `f8ea2f1a2`). F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active`, fresh CodeRabbit review still not posted at 08:54Z (~24 min past the 08:30Z re-queue; prior two CHANGES_REQUESTED `591805850`/`8d0d44bc3`). Re-trigger queue (serialized, one push per verdict, ready-chain order): **F7** (F4 3rd `916900f6a` + F6 `f8ea2f1a2` now in flight; F7 = PR #1567 head `b8b6ef39c` branch `vps2/f7-viewstate-loaded`). Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer`. Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `8fc4f7609` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `39423f27e` / F4 `916900f6a` / F5 `5d0a09b89` / F6 `f8ea2f1a2` / F7 `b8b6ef39c`.
Round 25 (2026-09-07 ~09:00Z): F4 #1562 3rd e2e FAILURE (run 34101891209 on `916900f6a`, 08:41:02Z→08:57:52Z ~16.8 min) = **72/13 overload class** — `Error: 13 tests failed.` with 7 provider timeouts (4×30s + 3×180s) + the same 6 probe-tag assertions (Bedrock user-agent, 2× thinking=disabled, probe-tag capture, xAI fast non-reasoning, xAI fast reasoning) — byte-identical failure set to F3's 6th (72/13) on the same new main; mutation-diff advisory failure (non-gate, 0-line content); unit tests not re-run to green on this head before e2e failure. 3 confirmed F4 failures (1st 83/2, 2nd 83/2, 3rd 72/13). Content-exonerated (same argument as round 24: F5/F0 heads e2e green on same main). Pool → 1 (only F6 in flight) → pushed F7 2nd no-op re-trigger `b8b6ef39c`→`553abe711` (09:00Z, ready-chain order; husky check-types all cache hits). Post-push sweep: all 10 PRs MERGEABLE (#1567 head `553abe711`). F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active`, fresh CodeRabbit review still not posted at 08:58Z (~28 min past the 08:30Z re-queue; timeline last event 08:28:42Z; prior two CHANGES_REQUESTED). E2E in flight: F6 `f8ea2f1a2` (run 34103168124, job 08:55:20Z). Re-trigger queue (serialized, one push per verdict, ready-chain order): **empty** — F6 + F7 both in flight; next push only after one of them lands (would be F6 or F7 2nd attempt depending on which fails). Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer`. Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `8fc4f7609` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `39423f27e` / F4 `916900f6a` / F5 `5d0a09b89` / F6 `f8ea2f1a2` / F7 `553abe711`.
Round 26 (2026-09-07 ~09:05Z): F6 #1565 2nd e2e FAILURE (run 34103168124 on `f8ea2f1a2`, 08:55:20Z→09:01:25Z ~6 min) = **87/1 class** — `87 passing (4m), 1 failing`: `Roo Code View State > three panels keep follow-up option mode switches isolated across ten staggered rounds` (`view-state.test.js:234`) — `Error: Timed out after 1 coordinated rounds; mode event counts: A:0, B:0, C:3; pending suggestions: 1` — same single-test timed-out signature as F6's prior 87/1 failure; provider suites (OpenRouter/xAI/GLM) all passing in the same run ⇒ load flake in the cross-instance coordination timing window, not a series-content regression (F6 head is the viewstate-feature head; the failing assertion is the staggered-rounds coordination probe, not a feature assertion). 2 confirmed F6 failures (both 87/1). Pool → 1 (only F7 in flight) → pushed F6 3rd no-op re-trigger `f8ea2f1a2`→`6e0b03ad6` (09:05Z; husky check-types all cache hits). Post-push sweep: all 10 PRs MERGEABLE (#1565 head `6e0b03ad6`). E2E in flight (pool 2/2): F6 `6e0b03ad6` (new run after 09:05Z) + F7 `553abe711` (run 34103431442, job 08:58:25Z). F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active`, fresh CodeRabbit review still not posted at 09:04Z (~34 min past the 08:30Z re-queue; prior two CHANGES_REQUESTED `591805850`/`8d0d44bc3`). Re-trigger queue (serialized, one push per verdict): empty until a verdict lands — next push = the failing PR's 3rd/4th attempt. Serial ready-marking still blocked at step 1: awaiting F0's fresh CodeRabbit APPROVED on `c51846007` → `awaiting-maintainer`. Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `8fc4f7609` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `39423f27e` / F4 `916900f6a` / F5 `5d0a09b89` / F6 `6e0b03ad6` / F7 `553abe711`.
Round 27 (2026-09-07 ~09:04Z, GOAL ROUND-CAP 60/60 — final state + handoff): no new e2e verdicts this round — F6 3rd (run 34103798295 on `6e0b03ad6`, job 09:02:34Z) and F7 2nd (run 34103431442 on `553abe711`, job 08:58:25Z) both in progress (logs not flushed; pool 2/2). F0 #1528 @ `c51846007`: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active`, fresh CodeRabbit review still not posted at 09:04Z (~34 min past the 08:30Z re-queue; timeline last event 08:28:42Z). Upstream main unchanged at `a3e31e14b` (02:25:32Z); all 10 heads verified merge-base `a3e31e14b` (round 23) ⇒ no rebase drift. Final sweep 10/10 MERGEABLE; F0 not-draft, F1a–F7 draft (serial queue intact). **Handoff (resume in this order):** (1) On F6/F7 e2e verdict: failure + pool ≤1 → push that PR's next no-op re-trigger from worktree `vps2-rebase` (fetch branch, `git commit-tree $tree -p $tip -m 'ci: re-trigger e2e (no-op)…'`, `git push --force-with-lease=refs/heads/<branch>:<tip> origin <new>:refs/heads/<branch>`), then 10/10 mergeable sweep + ledger append (this issue #41, sync via `gh issue edit 41 --repo easonLiangWorldedtech/Zoo-Code --body-file` + readback). (2) On F0 fresh CodeRabbit APPROVED on `c51846007` (→ `awaiting-maintainer`): `gh pr ready 1546 --repo Zoo-Code-Org/Zoo-Code`, then serial chain 1550→1552→1555→1560→1562→1557→1565→1567, each only after that PR reaches `awaiting-maintainer`; stop when F7 #1567 passes the bot. (3) DO NOT merge, DO NOT tag finished — await maintainer. E2E policy: max 2 concurrent, re-trigger only at pool ≤1, one push per verdict, no content/test changes for load flakes (classes 83/2 60s-timeout, 72/13 overload, 87/1 coordinated-rounds timeout are infra). F5 #1557 @ `5d0a09b89` e2e already green — no re-trigger. F0 advisory mutation-diff RED is permanent (7 survivors, 3 unkillable) — documented non-blocking.
Round 28 (2026-09-07 ~09:47Z): F6 3rd + F7 2nd e2e verdicts landed (both **87/1 class**, same test, identical CI signature `Timed out after 1 coordinated rounds; mode event counts: A:0, B:0, C:3; pending suggestions: 1`, view-state.test.js:234): F6 3rd run 34103798295 on `6e0b03ad6` (job 09:02:34Z→09:08:35Z, verdict 09:08:40Z); F7 2nd run 34103431442 on `553abe711` (job 08:58:25Z→09:04:46Z). Same test `Roo Code View State > three panels keep follow-up option mode switches isolated across ten staggered rounds` now **5/5 identical failures** (F6 1st–3rd `3d96e65f1`/`f8ea2f1a2`/`6e0b03ad6`; F7 1st–2nd `b8b6ef39c`/`553abe711`) — it is F6's own new test (not covered by F5 green @ `5d0a09b89`, which predates the test). 5/5 identical assertions crossed the local-repro decision boundary (≥2 same-assertion failures + failure on F6's own new code path) → **local repro on the quiet local machine at `6e0b03ad6` (worktree vps2-rebase, detached): REPRODUCED** — `USE_MOCK=true TEST_FILE=view-state.test pnpm --filter @roo-code/vscode-e2e test:run` → `2 passing (56s), 1 failing`, same test, local signature `Timed out after 0 coordinated rounds; mode event counts: A:0, B:0, C:0; pending suggestions: 1` (view-state.test.js:234; full log `local-e2e-viewstate.log` in worktree). Verdict: **not a CI-load flake** — the 3-panel follow-up coordination is timing-fragile on a quiet desktop too (background-tab webviews A/B do not deliver follow-up suggestions within the 30s budget; locally round 1 never even released, CI released round 1 and the focused panel C auto-processed 3). Old-series study (user-requested, "should be fixed before"): the 3-panel test is **byte-identical** between old #977 final head `e9a44b2fa` and current F6 head `6e0b03ad6` (diff = +88 lines only = F6's new 4th reload test); old-series commits show this suite's flake-fighting (`62682dfb4` poll-before-assert, `fab50e0a4` diagnostics for flaky viewStates read, `8ad5cdc77` 30s budget, `c518d7284` derive rounds from fixture) — but those fixed the persisted-viewStates READ wait, **not** this 3-panel coordinated wait; old #977 final e2e-mock run (33890940259, 2026-09-04T15:41Z) succeeded ONCE (lucky timing, same runner pool), and old sibling final heads #966/#979 show e2e FAILURE (1 recorded run each). So this test was never "fixed before" — the old pass was lucky timing; 0/6 recent runs (5 CI + 1 local) pass. - **Policy change needed — user decision required (no content/test changes without approval)**: (a) change the test (widen 30s budget and/or restructure the global-round coordination into per-panel polling) → new F6 commit + re-trigger; (b) product-level look at background-tab follow-up suggestion delivery; (c) keep re-triggering and hope (evidence: fragile, 0/6). F0 unchanged @ `c51846007`: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active`, no fresh CodeRabbit review as of 09:43Z (~73 min past 08:30Z re-queue; slow, no lever). Pool = 0; no push this round. No head changes this round. Upstream main unchanged @ `a3e31e14b`; all 10 PRs mergeable (last sweep 09:04Z).
Round 29 (2026-09-07 ~10:10Z, product-side investigation per user decision "先查產品面"): **ROOT CAUSE FOUND — the 87/1 failure is a deterministic product/test contract mismatch, not a flake.** Evidence chain: (1) CI log of run 34103798295 (F6 3rd) contains `suggestion delivery failures: <A-taskId>:A follow-up round 1, <B-taskId>:B follow-up round 1` — i.e. `selectTaskFollowupSuggestion` returned `delivered=false` for background tasks A and B at the first coordinated release: their task ids were already absent from the extension API's `tasksById` control registry. (2) `src/extension/api.ts` (F0 port `cfa678f1c`): `TaskUnfocused` handler → `removeRegisteredTask(task)` — an unfocused/evicted/aborted task is unregistered, so ask-controls return false for it. (3) `src/activate/registerCommands.ts` `openClineInNewTab` (new F0 chain commit `e6bc7a294` "fix(activate): target title-bar commands to their click-origin instance", ancestor of F0 head `c51846007`, **NOT present in old CS `e9a44b2fa`**): now REUSES the single tracked `tabPanel` on repeat calls ("a repeated 'Open in editor' click reveals the existing tab's panel") — so the 2nd/3rd `startNewTask({newTab:true})` returns the FIRST tab's provider. (4) `api.startNewTask` then calls `provider.evictCurrentTask()` (api.ts:215) on that shared provider → `removeClineFromStack` → `TaskUnfocused` + `abortTask(true)` + dispose → the previous task (A, then B) is ABORTED and unregistered. (5) Bonus defect: `startNewTask` re-calls `registerListeners(provider)` on the reused provider → N copies of the TaskCreated handler → the Nth task's events are re-emitted N times — CI's `C:3` = one real mode switch × 3 handler copies (A:1 copy, B:2, C:3). Both observed signatures fully explained by the same mechanism: CI = A/B emitted their first follow-up ask before eviction (round 1 released; A/B answers undeliverable; C switches once, triple-emitted; deadlock) → `Timed out after 1 coordinated rounds; A:0, B:0, C:3; pending suggestions: 1`; local = A/B evicted before their first aimock response (nothing pending for A/B) → `Timed out after 0 coordinated rounds; A:0, B:0, C:0; pending suggestions: 1`. (6) Old-series contrast: old CS `e9a44b2fa` `openClineInNewTab` created a FRESH webview panel + provider unconditionally (no tracking, no reuse, no cross-eviction) → the 3-panel test's premise (3 concurrent independent tab tasks) held in the old series and its final-head e2e pass (2026-09-04) ran against a working product. (7) The 3-panel test exists ONLY in F6/F7 heads (F4 `916900f6a`/F5 `5d0a09b89` lack `view-state.test.ts` entirely) ⇒ F1a–F5 e2e failures (83/2, 72/13) remain unrelated load flakes; F6/F7 87/1 = this defect. **Verdict: re-triggering cannot fix this (0/6 deterministic).** Fix options: (A) targeted product fix in F6's incremental diff — `startNewTask({newTab:true})` creates a FRESH tab (old-series behavior, matches the newTab contract) via a creation helper extracted from `openClineInNewTab`, UI button path keeps the reuse (F0's pinned creation-branch specs stay green) + duplicate-registration guard in `registerListeners`; F0–F5 heads untouched (their CI/bot state preserved); F6 re-push + F7 rebase + their CIs/reviews re-run; (B) fix in F0 itself + rebase F1a–F7 (9 PRs, chain CI/bot restart); (C) shrink the test to 2 panels (sidebar + 1 tab) — deviates from the old-CS test contract. **Awaiting user approval before any content/test change.**
Round 30 (2026-09-07 ~10:30Z, Option A implementation per user approval): root-cause fix shipped in F6's incremental diff. Changes (4 files, +107/-23): `src/activate/registerCommands.ts` — creation body extracted into a new exported `createClineTabPanel` (always creates a fresh panel + provider; its dispose handler clears the tracked ref only when it still points at the disposed panel); `openClineInNewTab` keeps F0's single-tab reuse block and delegates creation to the helper (UI button path unchanged, F0's pinned creation-branch specs stay green). `src/extension/api.ts` — `startNewTask({newTab:true})` now calls `createClineTabPanel` (the `evictCurrentTask()` lands on a fresh provider instead of aborting the task the tracked tab is already serving); `registerListeners` is guarded by a per-provider `Set` (duplicate registration would stack TaskCreated handler copies → Nth task events re-emitted N×, the CI `C:3` signature). Specs: `api-task-control.spec.ts` mock factory/rename only (assertions verbatim: evictCurrentTask once, createTask args, `{context, outputChannel}` call); +1 new `createClineTabPanel` unit test pinning the never-reuse contract (tracked tab present → no reveal, `createWebviewPanel` ×1). Gates: vitest 50/50 (both specs), `tsc --noEmit` clean, eslint `--prune-suppressions --max-warnings=0` exit 0 on all 4 files (suppression JSON semantically identical to HEAD — EOL-only churn reverted, counts flat), prettier clean (single fix: import line split, >120). **Local e2e single slot: 3/3 PASS (55s), exit 0** — the 0/6 deterministic 3-panel test now passes: `three panels keep follow-up option mode switches isolated across ten staggered rounds (29735ms)` ✔, zero delivery failures / timeouts in the log. F6 pushed (fast-forward, no force): `6e0b03ad6` → `d2d0a424e` ("fix(api): give startNewTask(newTab) a fresh tab panel instead of reusing the tracked one"); pre-push check-types 11/11. F7 rebased onto the new F6 head as a single content commit (its two no-op re-trigger commits were 0-line and tree-identical to the content commit, dropped): `553abe711` → `f4b20617c` ("fix(webview): show modes skeleton until view state loads"), `--force-with-lease` from `553abe711`. 10/10 MERGEABLE sweep post-push (F0 not-draft; F1a?F7 draft). CI: F6/F7 required checks (incl. e2e-mock) re-run on the new heads — first CI test of the fix; the local 3-panel run finished 29.7s of the 30s budget (passing but tight; if CI flakes on the same assertion under runner load, apply the no-op re-trigger lever once before considering any test change, and no test change without user approval). F0 #1528 @ `c51846007` unchanged: 6/6 required green, labels `awaiting-coderabbit`+`coderabbit-review-active`, no fresh CodeRabbit review as of ~10:18Z (~48 min past the 08:30Z re-queue; prior two CHANGES_REQUESTED on older heads `591805850`/`8d0d44bc3`). Serial ready-marking still blocked at step 1: awaiting F0 `awaiting-maintainer` (then `gh pr ready 1546` → chain 1550?552?555?560?562?557?565?567; stop at F7 pass; no merge, no finished tag). Head list: F0 `c51846007` / F1a `3c43a9e7c` / F1b `8fc4f7609` / F1c `18122c3c1` / F2 `f4621e82e` / F3 `39423f27e` / F4 `916900f6a` / F5 `5d0a09b89` / F6 `d2d0a424e` / F7 `f4b20617c`.

- **Content source of record (CS)**: `e9a44b2fa1be8c008ff36b927e580bca6282087b` (worktree `pr977-base`, #977 head) — #977 content is ported hunk-by-hunk from here.
- **Findings ledger**: `plans/viewstate-gap-review.md` — six-track DTE gap review (A1–A6): **0 blocker · 3 major · 14 minor · 10 nit**, all dispositioned into fix units F0–F7 or the parked-items register (12 items, all bounded).
- **Left open deliberately (by decision, 2026-09-05)**: upstream issue #568 (community Fork-Join feature request — the vps2 series is its state-isolation foundation, not its fulfillment) and upstream PR #926 (cross-instance task-history lock addressing #920 — task history is outside the per-view state surface; standalone fix track). F7 (#915 skeleton port, ~20 src + ~80 test) was un-parked on 2026-09-06 at user request and is queued last (wave 6, after F6).

## Unit status (maintained as PRs open/merge)

| Seq | Unit | Scope | Stack base | PR | a+d (measured) | State |
|-----|------|-------|------------|----|----------------|-------|
| 1 | **F0** | button targeting (D1–D4): no-wipe `setPanel`; click-origin sidebar title-bar targeting; 4 `*InTab` command ids; `ClineProvider.getInstanceForView`; tab-panel reuse; `focusInput` condition | main @ a3e31e14b (rebased) | [Zoo-Code-Org#1528](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1528) (ready for review 2026-09-07; issue [#1529](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1529)) | **484** (411+/73−) | code complete, pushed `vps2/f0-button-targeting` @ `8d0d44bc3` (CodeRabbit CHANGES_REQUESTED fix 2026-09-07 on `591805850` — serializes `openClineInNewTab` via shared in-flight promise + hardens view-identity test helpers, 4 files +123/−88); new-head CI 2026-09-07: all 6 required checks green on `8d0d44bc3` (check-translations, compile, knip, e2e-mock, platform-unit-test ubuntu+windows); `mutation-diff` FAILURE (advisory — not in the required 6 per ruleset; diff-scope widening from the IIFE wrapper, see round 15); CodeRabbit CHANGES_REQUESTED on `8d0d44bc3` 08:06:48Z (2 actionable: pin serialization results to mocked-constructor instance + 4 focused creation-branch tests) → fix commit `c51846007` pushed 08:15Z (spec +118/−4, 5 new tests, vitest 200/200, tsc/eslint clean); new-head CI: 5/6 required green (e2e-mock SUCCESS first try on `c51846007` — no flake; check-translations/compile/knip/unit-ubuntu green; unit-windows in progress 08:26Z); `mutation-diff` advisory FAILURE 38→7 surviving (4 of 7 behaviorally equivalent, unkillable without production changes — non-blocking, see round 17); fresh CodeRabbit review on `c51846007` in flight (labels `[]`) — awaiting APPROVED → `awaiting-maintainer` |
| 2 | **F1a** | view identity + durable core (F1 split 1/3): `ClineProvider` persistence primitives (write queue, prune, rekey, `setViewStateId` + `__proto__` guard, `loadViewState`, `saveViewState`), view-local buffer (`setValue`/`setValues`/`getValues` + mutation helpers), postMessage void-ack, resetState clears; types `viewStates` record; spec adapted into the existing `ClineProvider.spec.ts` fixture (viewId/save/load/prune-baseline + `__proto__` + ack + history-restore H5–H7; no mock-preamble duplication) | F0 head `e6bc7a294` (re-stacked) | [Zoo-Code-Org#1546](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1546) (draft; issue [#1547](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1547)) | **999** (976+/23−) | code complete, pushed `vps2/f1a-view-identity` @ `3c43a9e7c`; all gates green (179/179 mutants killed) |
| 3 | **F1b** | `getState` merge (F1 split 2/3): `mergedStateValues` + `getState()` body; spec: getState-merging + local-state-isolation describes (both assert via the merge) | F1a head `3c43a9e7c` | [Zoo-Code-Org#1550](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1550) (draft; issue [#1549](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1549)) | **717** (622+/95−) | code complete, pushed `vps2/f1b-getstate-merge` @ `e75806c44` (empty commits 2026-09-07: #4 `019c84dcf`→`6d13672b4`, #5 `6d13672b4`→`e75806c44`, #6 `e75806c44`→`e99726fbd`, #7 `e99726fbd`→`8fc4f7609` — e2e-mock 83/2 load flake, 7 confirmed failures (7th run 34101217142 08:40:36Z = 83/2 class, 08:33:32Z→08:40:36Z ~7.1 min), 0-line diff); all gates green (127/127 mutants killed; mutation-diff SUCCESS on `8fc4f7609`) |
| 4 | **F1c** | webview identity + launch (F1 split 3/3): WMH `webviewDidLaunch` → `setViewStateId` + launch-time re-pin via merged `getState()`, `updateSettings` via `provider.setValue`; webview-ui id-helper + `acquireVsCodeApi` fallback + new prod-branch `vsCodeApi` tests; WMH spec launch describe | F1b head `a37d0850e` | [Zoo-Code-Org#1552](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1552) (draft; issue [#1551](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1551)) | **545** (528+/17−) | code complete, pushed `vps2/f1c-webview-identity` @ `18122c3c1` (empty commits 2026-09-07: #4 `1c90a81b2`→`03c46f714`, #5 `03c46f714`→`acb18b2f9`, #6 `acb18b2f9`→`18122c3c1` — e2e-mock 83/2 load flake, 6 confirmed failures (6th run 34100554814 08:32Z = 83/2 class, 08:26:01Z→08:32:58Z ~6.9 min), 0-line diff); all gates green (stryker 59 killed / 0 survived / 6 ignored; unit tests both platforms green) |
| 5 | **F2** | durable viewStates: creates `ClineProvider.parallelMode.spec.ts` (588-line preamble + container fixture); pruning edge tests (missing `updatedAt`/ties) + #1065 50-entry-cap regression | F1c head `6021fee5d` | [Zoo-Code-Org#1555](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1555) (draft; issue [#1553](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1553); first-cut PR #1554 closed as superseded) | **733** (733+/0−) | code complete, pushed `vps2/f2-durable-viewstates` @ `8dd2a17ae` (empty commits 2026-09-07 on `3dcf4bfe3`: re-trigger #3, #4 `8dd2a17ae`→`f4621e82e` 08:29Z — e2e-mock load flake re-triggers; 4th e2e run 34100711329 FAILURE 08:34:46Z = 83/2 class (83 passing + same two 60s-timeout pair, ~7 min), 0-line diff); all gates green (3/3; stryker vacuous — 0 executable changed lines) |
| 6 | **F3** | mode/profile writes: webview delete-path via `deleteProviderProfile` + test; `loadViewState` dead-mode-pin validation + test; CP handleModeSwitch/upsert/activate/deleteProviderProfile hunks (incl. `repointPersistedViewStates` def) + `Task.ts`/`SwitchModeTool` + `sticky-mode` spec; handleModeSwitch durable-write tests H3/H4 (reclassified from F1 — base `updateGlobalState` bypasses public `setValue`); parallelMode F3 describes (CS L1364–1674) | F2 head `92b109654` | [Zoo-Code-Org#1560](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1560) (draft; issue [Zoo-Code-Org#1559](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1559)) | **984** (919+/65−, soft-over 400 — documented in PR body) | code complete, pushed `vps2/f3-mode-profile-writes` @ `39423f27e` (empty commits 2026-09-07: #2 on `f930a24ad` — e2e-mock load flake + windows unit-test turbo-cache infra flake, #6 `78f2a8296`→`39423f27e` 08:37Z — e2e-mock load flake, 5 confirmed failures (5th run 34097307741 08:07Z = 72/13 overload class), 0-line diff); all gates green (stryker 61/63 killed, 2 ignored as documented equivalent no-ops, 0 survived/uncovered) |
| 7 | **F4** | cross-instance reset + import invalidation: port #981 broadcast (single write-queue clear) + importExport hook; parallelMode multi-instance + `_clearViewLocalState` describes (CS L1675–1789, byte-identical, 5 tests); CP.spec forward fix (F3 resetState sentinel `toEqual({})` → `toBeUndefined()`) + gate-required multi-instance test; importExport skip-test console.warn negative assertion (gate-required) | F3 head `7d91ff4bd` | [Zoo-Code-Org#1562](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1562) (draft; issue [Zoo-Code-Org#1561](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1561)) | **343** (342+/1−) | code complete, pushed `vps2/f4-cross-instance-reset` @ `916900f6a` (empty commits 2026-09-07: #2 on `da97f98ee` — e2e-mock load flake, #3 `b6aea7e8d`→`916900f6a` 08:41Z — e2e-mock load flake, 2 confirmed failures, 0-line diff); all gates green (stryker 11/11 killed, 0 survived/uncovered; the 5 surviving mutants at the first gate head are killed by the two test additions, no directives) |
| 8 | **F5** | API wiring + task controls: `setConfiguration` doc fix; `preserveOpenTabs` @param; spec split; sidebar-only limitation documented | F4 head `da97f98ee` (re-stacked 2026-09-07 ×2: original commit `126a1fc4d` on the F1a head → F2 head `b5875118e` → F4 head `95a18b2f7`; the 5 file blobs are byte-identical across all three) | [Zoo-Code-Org#1557](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1557) (draft; issue [#1556](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1556)) | **557** (549+/8−, `95a18b2f7`) | code complete, pushed `vps2/f5-api-wiring` @ `5d0a09b89` (empty commit 2026-09-07 on `cfa678f1c` — windows unit-test turbo-cache flake re-trigger, 0-line diff), all gates green (vitest 17/17; stryker 42/42 killed, 0 survived) |
| 9 | **F6** | e2e suite (test-only): port/extend the view-state suite + A5 minor 2 (webview-reload/rehydration test) | F5 head `cfa678f1c` (re-stacked; the original plan of main @ post-F1a–F1c/F3 was obsoleted by the F3–F5 re-stack) | [Zoo-Code-Org#1565](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1565) (draft; issue [#1564](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1564)) | **543** (543+/0−) | code complete, pushed `vps2/f6-e2e-viewstate` @ `3d96e65f1` (empty commit 2026-09-07 on `09876aeea` — e2e-mock flake re-trigger, 0-line diff), all gates green (full solo e2e 2 runs — run 2: 71 passing / 5 failing, view-state 3/3 green by name; the 5 failures = pre-existing documented load-class timeouts, byte-identical across runs; stryker vacuous — 0 changed executable lines; the authored test carries the 2-line command-ID fix with bundle evidence) |
| 10 | **F7** | view-state loading skeleton (#915): ExtensionStateContext viewStateLoaded plumbing (interface/useState/state-handler flip/context value), ModesView skeleton early return, the folded App.tsx −3 (the bare webviewDidLaunch useEffect — the F1c-owned #928 hunk F1c shipped without), App.spec getViewStateId mock (+1, absent at base), spec tests (skeleton + 3 context lifecycle tests) | F6 head `09876aeea` | [Zoo-Code-Org#1567](https://github.com/Zoo-Code-Org/Zoo-Code/pull/1567) (draft; issue [#1566](https://github.com/Zoo-Code-Org/Zoo-Code/issues/1566)) | **101** (96+/5−) | code complete, pushed `vps2/f7-viewstate-loaded` @ `b8b6ef39c` (empty commit 2026-09-07 on `ac364cb20` — e2e-mock flake re-trigger, 0-line diff), all gates green (vitest 56/56 incl. the F1c webviewDidLaunch payload sentinel; check-types/eslint/prettier clean, suppressions flat at 346; stryker non-vacuous 7/7 killed / 0 survived, independently re-verified by coordinator re-run) |

F6 note (2026-09-07): the authored reload test first failed deterministically on workbench.action.toggleLeftSidebarVisibility (command-not-found; that ID has 0 occurrences in the pinned VS Code 1.100.0 workbench bundle, while workbench.action.toggleSidebarVisibility is present) — fixed by swapping the 2 command strings only (assertions byte-identical before/after, verified by reconstruction against the CS suite + 87-line authored span); run 1: 70 passing / 6 failing (the 5 load-class + the command bug), run 2: 71 passing / 5 failing (the same 5 load-class, byte-identical: execute_command simple-echo 60s / custom-working-directory 60s / multiple-sequentially 90s / long-running 60s + use_mcp_tool before-all 30s, all out/suite/utils.js:15, in suites this diff does not touch; the class is documented at run A: 7 concurrent-load timeouts). Both full logs kept as evidence of record (f6w-e2e.log / f6w-e2e-run2.log).

F7 note (2026-09-07): the worker's first stryker-diff run used the worktree's tracked (upstream) scripts/stryker-diff.mjs, which fails on win32 (spawnSync node_modules/.bin/vitest ENOENT — Node on Windows cannot exec pnpm's extensionless .bin shims); the win32 shim copy from pr977-base (never committed) re-ran green — 16 executable changed lines, 7 raw mutants (ModesView.tsx L617 ×5, ExtensionStateContext.tsx L294 + L351 ×2), 7/7 killed, 0 survived / no-coverage / timeout (thresholds high=low=100). The coordinator independently re-ran the same gate (exit 0) and kept the reports as evidence of record (f7c-stryker.log / f7c-stryker-reports/). The only non-verbatim port change: apiProvider "anthropic" to providerIdentifiers.anthropic in the new context spec test (fork lint rule zoo/no-raw-provider-identifiers; the import was already present).

## Severity rollup (six-track DTE gap review, 2026-09-05)

| Track | Scope | blocker | major | minor | nit | Verdict |
|-------|-------|---------|-------|-------|-----|---------|
| A1 | identity + persistence (CS review) | 0 | 0 | 4 | 2 | backbone sound; 3 small in-unit fixes |
| A2 | merge + write-path (CS review) | 0 | 1 | 3 | 1 | merge core sound; 2 write-path gaps |
| A3 | lifecycle (CS review) | 0 | 1 | 2 | 4 | additive port from #981 |
| A4 | API + e2e (CS review) | 0 | 1 | 2 | 2 | contract sound; one documented limitation |
| A5 | #929/F1–F5 series re-review (CS `e9a44b2fa`) | 0 | 0 | 2 | 1 | no blockers; e2e reload assertion → F6 |
| A6 | button-targeting audit (base `0d937c050`) | 0 | 4\* | 0 | 0 | D1–D4 confirmed; **fixed by F0** |

\* D1 `setPanel` mutual wipe · D2 no tab-panel reuse · D3 surface-blind visible instance (title-bar commands contribute to **both** `view/title` and `editor/title`; `newTask` is palette-only) · D4 `focusPanel` tab-first (self-resolves after F0's no-wipe — `focusPanel.ts` is never changed). A6 is the Phase-0 audit that produced F0.

**The three A1–A4 majors are bounded and additive:** (1) A2 — webview profile delete skips durable pin repair → F3; (2) A3 — resetState/import leave live siblings stale → F4; (3) A4 — API always targets the sidebar provider → parked (architectural), documented in F5.

## Phase 0 audit ledger (A5 re-review + A6 button audit)

**A6 button-targeting audit** (agent 8fb61842, base `0d937c050`):
- D1–D4 confirmed 4/4 at `registerCommands.ts` / `ClineProvider.ts` / `package.json` (D3 with correction: the five title-bar commands are contributed to **both** title bars — `view/title` L215-241 + `editor/title` L242-268; surface-blind `getVisibleInstance` returns insertion-order last-visible).
- 14-command surface matrix: ok 2 · D1 1 · D2 2 · D3 8 · D4 1.
- Pre-F0 spec: 3 describes / 15 its / **zero** multi-instance assertions (why the bug shipped).
- Alternate minimal-fix design (public `renderContext` / `getSurfaceProviderOrLog`) considered and **not adopted** — the plan-approved design (no-wipe + click-origin targeting + `*InTab` ids) is what is committed.

**A5 re-review** (agent bf591d91, CS `e9a44b2fa`): blocker 0 · major 0 · minor 2 · nit 1.
- Minor 1: #915 flash window remains — plumbing is correct (first "state" post awaits the per-view `setViewStateId` load before `postStateToWebview`); the flash is to `createInitialExtensionState` defaults (`ExtensionStateContext.tsx:200-224`), not another tab's state → **park (F7)**; fixed by the skeleton port.
- Minor 2: e2e lacks reload/rehydration assertion (`view-state.test.ts:41-308` never reloads; durable rehydration unit-covered only at `parallelMode.spec.ts:912-928`) → **fix-in-unit: one webview-reload e2e test → F6**.
- Nit: fixed `await sleep(1_000)` stagger (`view-state.test.ts:229`) → park (house pattern).
- #929 re-verification: **5/5 findings stale or superseded**.
- Already-satisfied (verified, no action): SettingsView fully `cachedState`-bound (#914 gate intact) · i18n 0 new strings · eslint suppressions down-only (ClineProvider 12→11, sticky-mode 37→33) · dispose retains the durable record · resetState clears · launch plumbing order.

## Risk flags (quick reference)

- **Budget**: F0 = **484 a+d measured** — over the 400 soft by +112 (mutation-killing tests: per-post distinct-error catch pinning, awaited no-op matrix, real-class `getInstanceForView` coverage); under the 600 hard cap; composition note in the PR body.
- **F1 mini-stack (decision 2026-09-05)**: F1 measured ≈ 2585 a+d against the CS port (2.6× the 1000 hard cap) ⇒ F1 → F1a/F1b/F1c. F1a → F1b → F1c is content-dependent (fields → getState merge → webview re-pin) and serial; F1a ≈ 950 est is the only unit near the 1000 hard cap — measure at cut, split lever = largest impl-hunk group (never tests). The CS F1 spec (1790-line `parallelMode.spec.ts`) is NOT ported as one file: the F1-series describes are rewritten into the existing `ClineProvider.spec.ts` fixture (drops the 588-line mock preamble); F2 creates `parallelMode.spec.ts` (preamble + pruning-edge + #1065 50-cap).
- **F4 port**: replaces #981's per-instance `setValue("viewStates", undefined)` with a **single write-queue clear** (queue at ClineProvider.ts:577) — must not regress the 50-entry prune cap (F2) or the no-secrets guarantee.
- **F5**: sidebar-only API targeting (A4 major) is **parked as architectural** (a per-view API handle would be a redesign); the PR body documents it.
- **F6**: the e2e reload/rehydration test (A5 minor 2) lands with the suite port; `test:ci:mock` gate required.
- **F7 parked**: the #915 flash remains until the skeleton port (~20 src + ~80 test; CS already has `viewStateId` message + `getViewStateId`/`createViewStateId` + `webviewDidLaunch` payload ⇒ #928's vscode.ts + webviewDidLaunch hunks are redundant).
- **Parked register**: 12 items (below) — all bounded, all documented; none needs rework.

---

# VPS2 Recreation Plan — durable per-view state as independent fixes

**This is the single execution source of truth.** It supersedes the closed #977 draft series (fork PR #34; upstream PRs #966 #928 #979 #981 #983) and `plans/pr-split-parallel-view-state.md` (old 4-PR inventory — historical reference for file-level attribution + commit table only).

**Objective**: recreate the durable per-view state fix/feat (superseded PR #977 + closed draft PRs/issues) as **a list of independent fixes (vps2)**: each PR is standalone, based on the same upstream base, gated by the standard skill gate suite, and carries its own upstream issue (created at PR-open time). This issue is the single coordination point.

## Independence guarantee

- **Every PR base = `upstream/main` `0d937c050810ebc305b34368bf6d6ae8c1f89966`** (the provider-files-only upstream commit; zero core-file impact).
- Stale sibling heads (`e7512a08b` #981, `9a246c8e5` #928, `e21ff41e9` #979, `2d162d6dd` #983, `beaf30b90` #966) are **porting sources only, never bases**.
- Content source of record for what-the-feature-should-be: CS `e9a44b2fa1be8c008ff36b927e580bca6282087b` (worktree `pr977-base`).
- Merge order (content dependencies): F0 ∥ F1a → F1b → (F1c ∥ F2) → F3 → F4 → F5 → F6. The F1 mini-stack (F1a → F1b → F1c) is content-dependent and serial — each PR rebases onto `main` after its predecessor merges; F1c and F2 are independent of each other. F5 unblocks once F1a/F2 merge; F6 once F1a–F1c + F3 merge.

## Fix table

| Fix | Unit | Scope (in-unit landings) | a+d (est / measured) | Deps |
|-----|------|--------------------------|----------------------|------|
| **F0** | button targeting | D1–D4: no-wipe `setPanel`; sidebar title-bar handlers target the click-origin `provider`; 4 new `*InTab` command ids (packages/types/src/vscode.ts) + editor/title rewire (zero new i18n strings); `ClineProvider.getInstanceForView`; tab-panel reuse in `openClineInNewTab`; `focusInput` condition `sidebarPanel && !tabPanel`; multi-instance spec rewrite | **484 measured** (411+/73−, `171620621`) | — |
| **F1a** | per-view identity + durable core | CP persistence primitives (write queue, prune, rekey, `setViewStateId` + `__proto__` guard, `loadViewState`, `saveViewState`) + view-local buffer (`setValue`/`setValues`/`getValues` + `_save/_update/_persist/_clearViewLocalState` helpers) + postMessage void-ack + resetState clears; types `viewStates` record (global-settings +10 / vscode-extension-host +1 / index.test +5); spec adapted into the existing `ClineProvider.spec.ts` fixture (viewId + saveViewState + loadViewState + prune-baseline describes, `__proto__` test, postMessage ack test, history-restore H5–H7, `hasLoadedFullDetails` Once fix — CS F1 spec NOT ported as one file, no 588-line preamble duplication) | **999 measured** (976+/23−, `0a8ffc9e1`) | — |
| **F1b** | `getState` merge | `mergedStateValues` + `getState()` body (CS h17/h18); spec: getState-merging describe (CS L1022–1363) + local-state-isolation describe (CS L719–765 — asserts via the `getState()` merge, so it lands here, not F1a) | **717 measured** (622+/95−, `43b52aa11`) | F1a |
| **F1c** | webview identity + launch | WMH `webviewDidLaunch` → `setViewStateId` + launch-time re-pin via merged `getState()`; `updateSettings` via `provider.setValue`; webview-ui id-helper + `acquireVsCodeApi` fallback + prod-branch `setState`/`getState` try/catch; new prod-branch vscode tests; WMH spec launch describe (excl. mojibake comment hunk + unused `defaultModeSlug` import — F3 re-adds it) | **545 measured** (528+/17−, `090d2c87e`) | F1b |
| **F2** | durable viewStates | creates `ClineProvider.parallelMode.spec.ts` (588-line mock preamble + container fixture from CS); pruning edge tests (missing `updatedAt`/ties) + #1065 50-entry-cap regression | **733 measured** (733+/0−, `aa0f3b1b7`) | F1a, F1c |
| **F3** | mode/profile writes | webview `deleteApiConfiguration` routed through `deleteProviderProfile` (repoint call) + webview-delete-path spec test; `loadViewState` dead-mode-pin validation with `defaultModeSlug` fallback + dead-mode-pin spec test; CP handleModeSwitch/upsert/activate/deleteProviderProfile hunks (incl. `repointPersistedViewStates` def + `delegateParentAndOpenChild` `as any` removal); `Task.ts`/`SwitchModeTool` + specs; `sticky-mode` spec (+104); handleModeSwitch durable-write tests H3/H4 (reclassified from F1 — base `updateGlobalState` bypasses public `setValue`, so the durable write only exists via F3's explicit `saveViewState`); parallelMode F3 describes (profile mutations / activation / handleModeSwitch, CS L1364–1674) | ~500 est | F1b, F2 |
| **F4** | cross-instance reset + import invalidation | port #981 (`e7512a08b`): `broadcastResetToAllInstances` (ClineProvider.ts:3064, 12 lines) + resetState call site (:3108) + importExport hook (importExport.ts:39 + :393-394) — replace #981's per-instance `setValue("viewStates", undefined)` with a single write-queue clear (queue at ClineProvider.ts:577); no webviewMessageHandler wrapper (import case passes provider structurally at webviewMessageHandler.ts:1027); tests: port parallelMode.spec L1247-1357 (~110) + importExport.spec (+152) + parallelMode multi-instance + `_clearViewLocalState` describes (CS L1675–1791) | ~300 est | F2, F3 |
| **F5** | API wiring + task controls | stale setConfiguration doc fix (packages/types/src/api.ts:116-119 → writes target the sidebar/extension-host view); `preserveOpenTabs` @param JSDoc (api.ts:20-32); spec split: api-task-control.spec.ts (L96-263 + L265-305), api-set-configuration.spec.ts (happy path), api-configuration.spec.ts (getConfiguration flatten + strip-secrets, L70-102 new); PR body documents the sidebar-only API limitation (A4 major, parked) | **557 measured** (549+/8−, `b5875118e` re-stacked on F2 head `aa0f3b1b7`; soft-over — the #982 wiring-exists verdict is NOT HELD at the F1a head, so the CS `src/extension/api.ts` port is included; rationale in the PR body) | F1a, F2 |
| **F6** | e2e suite (test-only) | port/extend apps/vscode-e2e/src/suite/view-state.test.ts for the vps2 split; A5 minor 2: one webview-reload/rehydration e2e test | measured at cut | F1a–F1c, F3 |
| **F7** | #915 skeleton | skeleton port (~20 src + ~80 test) | ~100 est | F6 | **parked** |

**Budgets (binding, per the `zoo-code-coding-workflow` skill)**: ≤ 400 a+d per PR soft (`git diff --shortstat <stack-base> <branch>`, re-measure post-commit), **1000 hard cap — never ship a PR above 1000**; a unit over 1000 is split at its reserved point (largest test-file describe or largest impl-hunk group — tests travel with their code); per mutation-tested package: changed executable lines ≤ 500 and raw mutants ≤ 400 (stryker-diff manifest; test/spec/visual files are excluded from the manifest); the CS `parallelMode.spec.ts` (+1790) is NOT ported as one file — the F1-series describes are rewritten into the existing `ClineProvider.spec.ts` fixture (drops the 588-line mock preamble), F2 creates `parallelMode.spec.ts` (preamble + pruning-edge + #1065 50-cap), F3/F4 add their describes (CS L1364–1674 / L1675–1791); never edit `scripts/stryker-diff.mjs` in a commit (local win32 shim allowed, always `git checkout -- scripts/stryker-diff.mjs` before push).

## Per-fix workflow (identical for every F)

1. **Cut**: `git worktree add -b vps2/<slug> <path> 0d937c050810ebc305b34368bf6d6ae8c1f89966`; `pnpm install --frozen-lockfile`.
2. **Implement** from the gap-review in-unit landings; port hunks from the stale sibling heads by hand (diff context, not cherry-pick).
3. **Gates** (per `zoo-code-coding-workflow` skill):
   - per touched file: `pnpm --dir src exec eslint --prune-suppressions --max-warnings=0 <relative-file>` — touched-file suppression counts must not rise (prune re-indents the json; check counts, revert if unchanged);
   - `pnpm check-types`;
   - targeted vitest per touched package (`pnpm --dir src exec vitest run <spec>`, `pnpm --dir packages/types exec vitest run <spec>`, etc.);
   - `node scripts/find-missing-translations.js` only if locales touched;
   - e2e touched ⇒ `pnpm --filter @roo-code/vscode-e2e test:ci:mock`;
   - visual baselines only for webview-ui changes;
   - mutation preflight: verify changed executable lines in mutation-tested packages ≤ ~300 (manifest dry-run);
   - `node scripts/stryker-diff.mjs ci --base 0d937c050... --head <head>` (install local win32 shim first; revert shim before push).
4. **Budget check**: `git diff --shortstat <stack-base> <branch>` ≤ 400 soft / 1000 hard (re-measure post-commit; over 1000 ⇒ split at the reserved point — never squeeze in).
5. **Commit**: one logical conventional commit, squashed; explicit `git add <paths>` only; `git status --short` before commit.
6. **Push**: revert the shim (`git checkout -- scripts/stryker-diff.mjs`) — push must contain only the intended files.
7. **Issue at PR-open time**: create the NEW upstream issue (bug report / feat description, self-contained) on Zoo-Code-Org/Zoo-Code.
8. **Draft PR**: base = upstream/main (no-wait stack: the displayed vs-main diff includes lower units until they merge — the body states the standalone a+d measured against the stack base, step 4); body references this tracking issue (cross-repo link) — **never** `Closes #xxx`; body carries the a+d measurement, gate evidence, parked-limitation notes where applicable.
9. **Bookkeeping**: update the unit-status table row above (PR link, a+d, state) after opening.

## Execution waves (max 3 parallel workers)

- **No-wait series policy (2026-09-06, user directive)**: cut each next unit immediately on the previous unit's pushed head — no merge wait. Every PR targets upstream `main` (displayed diff includes lower stack units until they merge; the body reports standalone a+d vs the stack base + local gate evidence; CI mutation-diff runs on the full stack diff until lower PRs merge). Rebase onto `main` after lower PRs merge = maintenance, never a blocker. Chain order: F1a → F1b → F1c → F2 → F3 → F4 → F5 → F6 → F7.

- **Wave 1 (done 2026-09-06)**: F0 (PR #1528 ∥ issue #1529) ∥ F1a (PR #1546 ∥ issue #1547) — both pushed, all gates green (F0 72/72; F1a 179/179 mutants killed).
- **Wave 2 (done 2026-09-07)**: F1b (PR #1550 ∥ issue #1549) — cut on F1a head `0a8ffc9e1` (no merge wait); all gates green (127/127 mutants killed, 0 survived); a+d 717 (spec-heavy unit — over the 400 soft, under the 1000 hard; composition note in the PR body).
- **Wave 3**: F1c (done — PR #1552 ∥ issue #1551, stacked on F1b head `43b52aa11`, a+d 545, all gates green) ∥ F2 (done — PR #1555 ∥ issue #1553, stacked on F1c head `090d2c87e`, a+d 733 test-only, all gates green (3/3; stryker vacuous); first-cut PR #1554 closed as superseded). F5 may join once F1a/F2 are in the stack (max 3 parallel workers).
- **Wave 4**: F3 (stacked on F2) → F4 (stacked on F3; its spec ports on top of F2's parallelMode spec).
- **Wave 3–4 (done 2026-09-07)**: F5 (PR #1557 ∥ issue #1556) — re-stacked on the F2 head `aa0f3b1b7` as `b5875118e` (the original commit `126a1fc4d` on the F1a head cherry-picked clean — F1b/F1c/F2 touch none of the 5 F5 files); a+d 557 (soft-over vs ~250 est — the #982 wiring-exists verdict is not held at the F1a head, so the CS `src/extension/api.ts` port is included; rationale in the PR body); all gates green (vitest 17/17; stryker 42/42 killed, 0 survived; coordinator re-verified independently).
- **Wave 4 in-flight (2026-09-07)**: F3 — worker commit `bd9ddccba` (11 files 625+/120−, stryker RED: 19 survived + 6 uncovered) is being corrected in place (branch unpushed): (a) WMH.ts reverted to base — the 18/12 hunk is the CS-only kimi-code OAuth divergence (register item 1; not in #979; its 3 mutants unkillable in scope, no suppressions); (b) WMH.spec keeps only #979's 4 mock fields + defaultModeSlug import (5/0 vs base) and restores the 3 stack-side pieces (legacy-repair test, Key-aware getValue mock, em-dash comment — register item 4 mojibake must not land); (c) CP.ts `setViewStateId` __proto__ guard restored — the residual's guard removal is lineage divergence (F1a hardening), not F3 content (without it viewStateId="__proto__" corrupts the viewStates map prototype and the view's own writes are silently dropped); (d) CP.spec `describe("setViewStateId")` block restored (acceptance = exactly the 2 assertion-rewrite hunks); 7 remaining non-WMH mutants (SwitchModeTool L62; CP L638/L641/L648/L655/L650/L651/L1999) killed by spec strengthening. Upstream mapping: issue #978 = F3; PR #979 = port source (stale head e21ff41e9); #979's mode-routing WMH.spec describe exists in neither CS nor stack — ported by no unit (open question, logged).
- **F5 re-stack decision (2026-09-07)**: F5's parent is the F2 head `aa0f3b1b7`, which lacks the F3/F4 content F6's e2e suite needs (F6 RED root cause: at F5 head `handleModeSwitchUnlocked` does a global-only `updateGlobalState("mode", newMode)` and the sidebar's `viewLocalState.mode` = "code" masks global "ask" → deterministic `'code' !== 'ask'` at view-state.test.ts:90; F3's per-view `saveViewState("mode", newMode)` write is the missing production behavior). Fix: cherry-pick `b5875118e` onto the F4 head (5 files, zero overlap with F3/F4 → clean), force-push `vps2/f5-api-wiring`, update PR #1557 body (new head/parent/budget) + the F5 row with the new head SHA.
- **F6 STOP (2026-09-07)**: e2e gate RED on the F5 head (Run A concurrent ~12m: 67 pass/8 fail — 7 load-concurrent timeouts unverified + 1 real = CS test 1; Run B clean solo: test 1 deterministic failure both runs, test 2 timing-sensitive — passed under load). 4/4 CS blobs verified byte-identical and preserved uncommitted in `vps2-f6-wt`; reload test designed (sidebar hide/reshow via workbench.action.toggleLeftSidebarVisibility, 2s gaps + 5s settle; assertions (a) durable viewStates mode entry + (b) getConfiguration().mode; (c) skipped — no deterministic extension-side task-mode getter). Re-launches on the re-stacked F5 head as a fresh subagent (original worker d986ea52 no longer messageable). Windows quirk: `test:ci:mock`'s POSIX env prefix `USE_MOCK=true pnpm test:run` is broken (`'USE_MOCK' is not recognized`) → run `$env:USE_MOCK='true'; pnpm test:run` instead; flag in F6 PR body; script itself out of scope.
- **Wave 5**: F6 (stacked on the RE-STACKED F5 head — see F5 re-stack decision) — e2e suite, test-only.
- **Wave 6**: F7 (#915 skeleton port) — un-parked 2026-09-06 at user request; after F6.
- **Cleanup (after all units ship)**: the `pr977-base` shim stays the series shim source until then (each unit worktree copies it and reverts before its push); remove worktrees (`crfix-977`, `pr977-base`, `pr977-work`, `Zoo-Code-pr977`, then per-fix worktrees when done); delete stale local branches + `origin/feat|fix/*view*` remotes (keep `feat/view-local-state-base` until F6 merges); delete all `.tmp-*` incl. `.tmp-gh-token.txt`.

## PR body template (per fix PR)

```
Part of the vps2 durable per-view state series — tracked in easonLiangWorldedtech/Zoo-Code#41
(cross-repo: this PR is standalone against upstream/main @ 0d937c050).

## What
<2-3 sentences: which defect/gap, which fix unit (F#)>

## Design decisions
<bulleted, terse — from the gap-review ledger entries>

## Measurements
- a+d vs upstream/main @ 0d937c050: <N> (soft budget 400)
- src executable lines changed: <N> (mutation preflight)

## Gates
- eslint --prune-suppressions: pass (suppression counts unchanged: <...>)
- check-types: pass
- vitest: <suites> pass
- stryker-diff ci: <K killed / S survived (list) / 0 blocking>
- (e2e / i18n / visual: n/a or pass)

## Parked / documented
<limitations inherited from the gap review, e.g. F5 sidebar-only API targeting>

## Porting notes
<hunks ported from the stale sibling heads / CS (by id + line refs), all re-implemented against this base; plus the CS-hunks-not-ported register entries this PR observes (register in the tracking issue)>
```

## CS hunks intentionally NOT ported (verified 2026-09-05; observed by every vps2 PR)

1. **requestRouterModels kimi-code OAuth try/catch** (webviewMessageHandler prod hunk, WMH diff L246–281) + `routerModels.spec.ts` +29 test — out-of-scope error-handling change bundled in #977.
2. **`ApiConfigManager.tsx` className** `min-w-0 shrink` → `grow` (+2/−) — out-of-scope layout tweak.
3. **`ApiConfigManager.visual.tsx` deletion** (−36) + **2 PNG visual baselines** — visual-test rework bundled in #977, not a state change.
4. **Mojibake comment hunk** (WMH diff L160–167, `—?` corruption) — do not port.
5. **Unused `defaultModeSlug` import** in the WMH diff — dropped from F1c; F3 re-adds it with its dead-mode-pin test (the import lands with its first consumer).
6. **providers/\***, `.coderabbit.yaml`, `.github/*`, `CONTRIBUTING.md`, `.gitignore` — non-functional / repo-config churn.

## Parked-items register (12, all bounded)

1. Dev/prod viewStateId divergence (A1 minor 3) — inherent to browser mock.
2. Pre-launch write-queue/rekey orphan window + cross-session temp-id collision (A1 minor 2) — narrow window, prune-bounded, non-secret.
3. Cross-process write-queue interleave (A1 minor 4) — pre-existing memento semantics.
4. Redundant repoint branch (A1 nit) — cosmetic.
5. Launch-repair queue sync (A2 minor 3) — transient, self-healing.
6. Same-provider two-writer interleave test (A2 minor 4 tail) — cross-instance interleave already covered.
7. Flat-mutation apiConfiguration replace (A2 nit) — coherent via getState re-merge.
8. Editor-tab viewStates orphan after window reload (A3 minor 3) — no panel serializer; prune-bounded.
9. Pre-hydration state-post window (A3 nit) — UI severity unestablished.
10. Sidebar-only API targeting (A4 major) — architectural; documented limitation.
11. Public task-scoped ask controls (A4 minor 1) — trust model matches pre-existing public approveCurrentAsk.
12. set-configuration implementation-detail pin (A4 nit) — brittle if inlined; acceptable.

## Superseded-items disposition (24 items; 23 closed 2026-09-05, #1065 close permission-blocked)

| Item | Title | Covered by |
|------|-------|------------|
| fork PR #34 | Merge #977 (durable per-view state base) into the file-write-safety trial branch | F1 + F2 (the #977 content, re-cut as independent fixes) |
| upstream #967 | [BUG] Per-view state lost on extension reload — no durable persistence | F1 + F2 |
| upstream #984 | [BUG] Add durable per-view state persistence for parallel tabs | F1 + F2 |
| upstream #1065 | Preserve durable editor view state across provider disposal | F2 (50-entry-cap regression; dispose-retention already satisfied in CS) — close permission-blocked (403); pointer comment posted |
| upstream #978 | [BUG] Isolate parallel mode and provider profile writes | F3 |
| upstream #980 | [BUG] Invalidate per-view state after reset and settings import | F4 |
| upstream #982 | [BUG] Sync API setConfiguration through provider state wiring | F5 (sidebar-only limitation documented) |
| upstream #913 | [UX] Multi-tab mode isolation — follow-up issues from PR #909 | the series as a whole (F0–F6); individual findings dispositioned in the ledger above |
| upstream #915 | [UX] Flash of incorrect mode/apiConfig when opening new editor tab | F7 (skeleton port; un-parked 2026-09-06 — wave 6) |
| upstream #929 | [909] CodeRabbit Review Findings — View State Isolation Bugs | A5 re-review: 5/5 findings verified stale or superseded |
| upstream #916 | [UX] Add visual indicator for per-view mode isolation | out of vps2 scope (UX polish; no fix unit) |
| upstream #917 | [UX] Add save indicator for per-view API configuration changes | out of vps2 scope (UX polish; no fix unit) |
| upstream #918 | [UX] Wire SettingsView Save button to syncViewStateToGlobal | out of vps2 scope (SettingsView cachedState binding already satisfied per A5; not part of CS content) |
| upstream #919 | [UX] Add telemetry/tooltip for view-local state persistence limitation | out of vps2 scope (UX polish; no fix unit) |
| upstream #920 | [BUG] Concurrent task history updates from parallel tabs may cause lost entries | out of vps2 scope (task history is outside the per-view state surface) |
| upstream #922 | [BUG] Switching mode/apiConfig while parallel tools are running may corrupt task state | F3 (per-view mode/profile write isolation; unknown-mode hardening verified in the gap review) |
| upstream #923 | [ENHANCEMENT] Support Fork-Join parallel modes with per-view state isolation | out of vps2 scope (per-view isolation layer = F1–F6; Fork-Join orchestration is a separate feature) |
| upstream #924 | [ENHANCEMENT] Add "New Tab" button beside "Start New Task" for parallel mode discoverability | out of vps2 scope (UX discoverability polish; no fix unit) — same disposition as #916/#917/#919 |
| upstream #921 | [BUG] Parent-child task delegation across parallel tabs may lose state | out of vps2 scope (task delegation is outside the per-view state surface) |
| upstream PR #966 | fix(webview): isolate parallel provider view state | F1/F2 content (porting source, stale head `beaf30b90`) |
| upstream PR #928 | fix(webview): add loading skeleton for view state initialization | F7 (#915 skeleton port; un-parked 2026-09-06; its vscode.ts + webviewDidLaunch hunks redundant in CS) |
| upstream PR #979 | fix(webview): isolate parallel mode and provider profile writes | F3 content (porting source, stale head `e21ff41e9`) |
| upstream PR #981 | fix(webview): invalidate per-view state after reset and import | F4 content (porting source, stale head `e7512a08b`) |
| upstream PR #983 | fix(api): sync setConfiguration with provider state | F5 content (porting source, stale head `2d162d6dd`) |

Related items left open by decision (not superseded): #568 (community Fork-Join feature request — vps2 is its state-isolation foundation, not its fulfillment) · PR #926 (cross-instance task-history lock for #920 — task history is outside the per-view state surface; standalone track).

## Bookkeeping checklist

- [x] 23/24 superseded items closed with tracker-pointer comments (2026-09-05; includes #922/#923/#924 found on the open-list audit); #1065 pointer comment posted, close permission-blocked (403). #568 + PR #926 left open by decision.
- [ ] Every fix PR carries its own upstream issue (created at PR-open time); PR bodies reference this issue by cross-repo link — **never** `Closes #xxx`.
- [ ] Unit-status rows carry final PR #, head SHA, and measured a+d (each ≤ 400 soft, or the 1000 hard cap with a composition note).
- [ ] Content fidelity: per-file CS hunks ported (no per-PR-head-only shapes shipped); `git diff main..CS -- <file>` deltas match this ledger.
- [ ] All F0–F7 merged to upstream/main (F7 un-parked 2026-09-06 at user request).
- [ ] Cleanup done: worktrees, stale branches, `.tmp-*` files (incl. `.tmp-gh-token.txt`).

## Non-goals

- No `Closes #xxx` in any PR body (cross-repo closure is the manual close-all only).
- No changesets / changelog edits.
- No architectural redesign of the per-view model (gap review verdict: zero blockers, additive fixes only).
- F7 (#915 skeleton) un-parked 2026-09-06 at user request; queued last (wave 6).



Seq	Unit	Scope	Stack base	PR	a+d (measured)	State
1	F0	button targeting (D1–D4): no-wipe `setPanel`; click-origin sidebar title-bar targeting; 4 `*InTab` command ids; `ClineProvider.getInstanceForView`; tab-panel reuse; `focusInput` condition	main @ `a3e31e1` (rebased)	Zoo-Code-Org#1528 (ready for review 2026-09-07; issue #1529)	484 (411+/73−)	code complete, pushed `vps2/f0-button-targeting` @ `8d0d44bc3` (CodeRabbit CHANGES_REQUESTED fix 2026-09-07 on `591805850` — serializes `openClineInNewTab` via shared in-flight promise + hardens view-identity test helpers, 4 files +123/−88); new-head CI 2026-09-07: all 6 required checks green on `8d0d44bc3` (check-translations, compile, knip, e2e-mock, platform-unit-test ubuntu+windows); `mutation-diff` FAILURE (advisory — not in the required 6 per ruleset; diff-scope widening from the IIFE wrapper, see round 15); CodeRabbit CHANGES_REQUESTED on `8d0d44bc3` 08:06:48Z (2 actionable: pin serialization results to mocked-constructor instance + 4 focused creation-branch tests) → fix commit `c51846007` pushed 08:15Z (spec +118/−4, 5 new tests, vitest 200/200, tsc/eslint clean); new-head CI: 5/6 required green (e2e-mock SUCCESS first try on `c51846007` — no flake; check-translations/compile/knip/unit-ubuntu green; unit-windows in progress 08:26Z); `mutation-diff` advisory FAILURE 38→7 surviving (4 of 7 behaviorally equivalent, unkillable without production changes — non-blocking, see round 17); fresh CodeRabbit review on `c51846007` in flight (labels `[]`) — awaiting APPROVED → `awaiting-maintainer`
2	F1a	view identity + durable core (F1 split 1/3): `ClineProvider` persistence primitives (write queue, prune, rekey, `setViewStateId` + `__proto__` guard, `loadViewState`, `saveViewState`), view-local buffer (`setValue`/`setValues`/`getValues` + mutation helpers), postMessage void-ack, resetState clears; types `viewStates` record; spec adapted into the existing `ClineProvider.spec.ts` fixture (viewId/save/load/prune-baseline + `__proto__` + ack + history-restore H5–H7; no mock-preamble duplication)	F0 head `e6bc7a294` (re-stacked)	Zoo-Code-Org#1546 (draft; issue #1547)	999 (976+/23−)	code complete, pushed `vps2/f1a-view-identity` @ `3c43a9e7c`; all gates green (179/179 mutants killed)
3	F1b	`getState` merge (F1 split 2/3): `mergedStateValues` + `getState()` body; spec: getState-merging + local-state-isolation describes (both assert via the merge)	F1a head `3c43a9e7c`	Zoo-Code-Org#1550 (draft; issue #1549)	717 (622+/95−)	code complete, pushed `vps2/f1b-getstate-merge` @ `e75806c44` (empty commits 2026-09-07: #4 `019c84dcf`→`6d13672b4`, #5 `6d13672b4`→`e75806c44`, #6 `e75806c44`→`e99726fbd`, #7 `e99726fbd`→`8fc4f7609` — e2e-mock 83/2 load flake, 7 confirmed failures (7th run 34101217142 08:40:36Z = 83/2 class, 08:33:32Z→08:40:36Z ~7.1 min), 0-line diff); all gates green (127/127 mutants killed; mutation-diff SUCCESS on `8fc4f7609`)
4	F1c	webview identity + launch (F1 split 3/3): WMH `webviewDidLaunch` → `setViewStateId` + launch-time re-pin via merged `getState()`, `updateSettings` via `provider.setValue`; webview-ui id-helper + `acquireVsCodeApi` fallback + new prod-branch `vsCodeApi` tests; WMH spec launch describe	F1b head `a37d0850e`	Zoo-Code-Org#1552 (draft; issue #1551)	545 (528+/17−)	code complete, pushed `vps2/f1c-webview-identity` @ `18122c3c1` (empty commits 2026-09-07: #4 `1c90a81b2`→`03c46f714`, #5 `03c46f714`→`acb18b2f9`, #6 `acb18b2f9`→`18122c3c1` — e2e-mock 83/2 load flake, 6 confirmed failures (6th run 34100554814 08:32Z = 83/2 class, 08:26:01Z→08:32:58Z ~6.9 min), 0-line diff); all gates green (stryker 59 killed / 0 survived / 6 ignored; unit tests both platforms green)
5	F2	durable viewStates: creates `ClineProvider.parallelMode.spec.ts` (588-line preamble + container fixture); pruning edge tests (missing `updatedAt`/ties) + Zoo-Code-Org#1065 50-entry-cap regression	F1c head `6021fee5d`	Zoo-Code-Org#1555 (draft; issue #1553; first-cut PR Zoo-Code-Org#1554 closed as superseded)	733 (733+/0−)	code complete, pushed `vps2/f2-durable-viewstates` @ `8dd2a17ae` (empty commits 2026-09-07 on `3dcf4bfe3`: re-trigger #3, #4 `8dd2a17ae`→`f4621e82e` 08:29Z — e2e-mock load flake re-triggers; 4th e2e run 34100711329 FAILURE 08:34:46Z = 83/2 class (83 passing + same two 60s-timeout pair, ~7 min), 0-line diff); all gates green (3/3; stryker vacuous — 0 executable changed lines)
6	F3	mode/profile writes: webview delete-path via `deleteProviderProfile` + test; `loadViewState` dead-mode-pin validation + test; CP handleModeSwitch/upsert/activate/deleteProviderProfile hunks (incl. `repointPersistedViewStates` def) + `Task.ts`/`SwitchModeTool` + `sticky-mode` spec; handleModeSwitch durable-write tests H3/H4 (reclassified from F1 — base `updateGlobalState` bypasses public `setValue`); parallelMode F3 describes (CS L1364–1674)	F2 head `92b109654`	Zoo-Code-Org#1560 (draft; issue Zoo-Code-Org#1559)	984 (919+/65−, soft-over 400 — documented in PR body)	code complete, pushed `vps2/f3-mode-profile-writes` @ `39423f27e` (empty commits 2026-09-07: #2 on `f930a24ad` — e2e-mock load flake + windows unit-test turbo-cache infra flake, #6 `78f2a8296`→`39423f27e` 08:37Z — e2e-mock load flake, 5 confirmed failures (5th run 34097307741 08:07Z = 72/13 overload class), 0-line diff); all gates green (stryker 61/63 killed, 2 ignored as documented equivalent no-ops, 0 survived/uncovered)
7	F4	cross-instance reset + import invalidation: port Zoo-Code-Org#981 broadcast (single write-queue clear) + importExport hook; parallelMode multi-instance + `_clearViewLocalState` describes (CS L1675–1789, byte-identical, 5 tests); CP.spec forward fix (F3 resetState sentinel `toEqual({})` → `toBeUndefined()`) + gate-required multi-instance test; importExport skip-test console.warn negative assertion (gate-required)	F3 head `7d91ff4bd`	Zoo-Code-Org#1562 (draft; issue Zoo-Code-Org#1561)	343 (342+/1−)	code complete, pushed `vps2/f4-cross-instance-reset` @ `916900f6a` (empty commits 2026-09-07: #2 on `da97f98ee` — e2e-mock load flake, #3 `b6aea7e8d`→`916900f6a` 08:41Z — e2e-mock load flake, 2 confirmed failures, 0-line diff); all gates green (stryker 11/11 killed, 0 survived/uncovered; the 5 surviving mutants at the first gate head are killed by the two test additions, no directives)
8	F5	API wiring + task controls: `setConfiguration` doc fix; `preserveOpenTabs` @PARAM; spec split; sidebar-only limitation documented	F4 head `da97f98ee` (re-stacked 2026-09-07 ×2: original commit `126a1fc4d` on the F1a head → F2 head `b5875118e` → F4 head `95a18b2f7`; the 5 file blobs are byte-identical across all three)	Zoo-Code-Org#1557 (draft; issue #1556)	557 (549+/8−, `95a18b2f7`)	code complete, pushed `vps2/f5-api-wiring` @ `5d0a09b89` (empty commit 2026-09-07 on `cfa678f1c` — windows unit-test turbo-cache flake re-trigger, 0-line diff), all gates green (vitest 17/17; stryker 42/42 killed, 0 survived)
9	F6	e2e suite (test-only): port/extend the view-state suite + A5 minor 2 (webview-reload/rehydration test)	F5 head `cfa678f1c` (re-stacked; the original plan of main @ post-F1a–F1c/F3 was obsoleted by the F3–F5 re-stack)	Zoo-Code-Org#1565 (draft; issue #1564)	543 (543+/0−)	code complete, pushed `vps2/f6-e2e-viewstate` @ `3d96e65f1` (empty commit 2026-09-07 on `09876aeea` — e2e-mock flake re-trigger, 0-line diff), all gates green (full solo e2e 2 runs — run 2: 71 passing / 5 failing, view-state 3/3 green by name; the 5 failures = pre-existing documented load-class timeouts, byte-identical across runs; stryker vacuous — 0 changed executable lines; the authored test carries the 2-line command-ID fix with bundle evidence)
10	F7	view-state loading skeleton (Zoo-Code-Org#915): ExtensionStateContext viewStateLoaded plumbing (interface/useState/state-handler flip/context value), ModesView skeleton early return, the folded App.tsx −3 (the bare webviewDidLaunch useEffect — the F1c-owned Zoo-Code-Org#928 hunk F1c shipped without), App.spec getViewStateId mock (+1, absent at base), spec tests (skeleton + 3 context lifecycle tests)	F6 head `09876aeea`	Zoo-Code-Org#1567 (draft; issue #1566)	101 (96+/5−)	code complete, pushed `vps2/f7-viewstate-loaded` @ `b8b6ef39c` (empty commit 2026-09-07 on `ac364cb20` — e2e-mock flake re-trigger, 0-line diff), all gates green (vitest 56/56 incl. the F1c webviewDidLaunch payload sentinel; check-types/eslint/prettier clean, suppressions flat at 346; stryker non-vacuous 7/7 killed / 0 survived, independently re-verified by coordinator re-run)

Navigation Menu

VPS2 durable per-view state - independent-fix series (supersedes #34 + the 21-item upstream draft series) #41

Unit status (maintained as PRs open/merge)

Severity rollup (six-track DTE gap review, 2026-09-05)

Phase 0 audit ledger (A5 re-review + A6 button audit)

Risk flags (quick reference)

VPS2 Recreation Plan — durable per-view state as independent fixes

Independence guarantee

Fix table

Per-fix workflow (identical for every F)

Execution waves (max 3 parallel workers)

PR body template (per fix PR)

CS hunks intentionally NOT ported (verified 2026-09-05; observed by every vps2 PR)

Parked-items register (12, all bounded)

Superseded-items disposition (24 items; 23 closed 2026-09-05, Zoo-Code-Org#1065 close permission-blocked)

Bookkeeping checklist

Non-goals

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Track	Scope	major	minor	nit	Verdict
A1	identity + persistence (CS review)	0	4	2	backbone sound; 3 small in-unit fixes
A2	merge + write-path (CS review)	1	3	1	merge core sound; 2 write-path gaps
A3	lifecycle (CS review)	1	2	4	additive port from Zoo-Code-Org#981
A4	API + e2e (CS review)	1	2	2	contract sound; one documented limitation
A5	Zoo-Code-Org#929/F1–F5 series re-review (CS `e9a44b2fa`)	0	2	1	no blockers; e2e reload assertion → F6
A6	button-targeting audit (base `0d937c050`)	4*	0	0	D1–D4 confirmed; fixed by F0

Fix	Unit	Scope (in-unit landings)	a+d (est / measured)	Deps
F0	button targeting	D1–D4: no-wipe `setPanel`; sidebar title-bar handlers target the click-origin `provider`; 4 new `*InTab` command ids (packages/types/src/vscode.ts) + editor/title rewire (zero new i18n strings); `ClineProvider.getInstanceForView`; tab-panel reuse in `openClineInNewTab`; `focusInput` condition `sidebarPanel && !tabPanel`; multi-instance spec rewrite	484 measured (411+/73−, `171620621`)	—
F1a	per-view identity + durable core	CP persistence primitives (write queue, prune, rekey, `setViewStateId` + `__proto__` guard, `loadViewState`, `saveViewState`) + view-local buffer (`setValue`/`setValues`/`getValues` + `_save/_update/_persist/_clearViewLocalState` helpers) + postMessage void-ack + resetState clears; types `viewStates` record (global-settings +10 / vscode-extension-host +1 / index.test +5); spec adapted into the existing `ClineProvider.spec.ts` fixture (viewId + saveViewState + loadViewState + prune-baseline describes, `__proto__` test, postMessage ack test, history-restore H5–H7, `hasLoadedFullDetails` Once fix — CS F1 spec NOT ported as one file, no 588-line preamble duplication)	999 measured (976+/23−, `0a8ffc9e1`)	—
F1b	`getState` merge	`mergedStateValues` + `getState()` body (CS h17/h18); spec: getState-merging describe (CS L1022–1363) + local-state-isolation describe (CS L719–765 — asserts via the `getState()` merge, so it lands here, not F1a)	717 measured (622+/95−, `43b52aa11`)	F1a
F1c	webview identity + launch	WMH `webviewDidLaunch` → `setViewStateId` + launch-time re-pin via merged `getState()`; `updateSettings` via `provider.setValue`; webview-ui id-helper + `acquireVsCodeApi` fallback + prod-branch `setState`/`getState` try/catch; new prod-branch vscode tests; WMH spec launch describe (excl. mojibake comment hunk + unused `defaultModeSlug` import — F3 re-adds it)	545 measured (528+/17−, `090d2c87e`)	F1b
F2	durable viewStates	creates `ClineProvider.parallelMode.spec.ts` (588-line mock preamble + container fixture from CS); pruning edge tests (missing `updatedAt`/ties) + Zoo-Code-Org#1065 50-entry-cap regression	733 measured (733+/0−, `aa0f3b1b7`)	F1a, F1c
F3	mode/profile writes	webview `deleteApiConfiguration` routed through `deleteProviderProfile` (repoint call) + webview-delete-path spec test; `loadViewState` dead-mode-pin validation with `defaultModeSlug` fallback + dead-mode-pin spec test; CP handleModeSwitch/upsert/activate/deleteProviderProfile hunks (incl. `repointPersistedViewStates` def + `delegateParentAndOpenChild` `as any` removal); `Task.ts`/`SwitchModeTool` + specs; `sticky-mode` spec (+104); handleModeSwitch durable-write tests H3/H4 (reclassified from F1 — base `updateGlobalState` bypasses public `setValue`, so the durable write only exists via F3's explicit `saveViewState`); parallelMode F3 describes (profile mutations / activation / handleModeSwitch, CS L1364–1674)	~500 est	F1b, F2
F4	cross-instance reset + import invalidation	port Zoo-Code-Org#981 (`e7512a08b`): `broadcastResetToAllInstances` (ClineProvider.ts:3064, 12 lines) + resetState call site (:3108) + importExport hook (importExport.ts:39 + :393-394) — replace Zoo-Code-Org#981's per-instance `setValue("viewStates", undefined)` with a single write-queue clear (queue at ClineProvider.ts:577); no webviewMessageHandler wrapper (import case passes provider structurally at webviewMessageHandler.ts:1027); tests: port parallelMode.spec L1247-1357 (~110) + importExport.spec (+152) + parallelMode multi-instance + `_clearViewLocalState` describes (CS L1675–1791)	~300 est	F2, F3
F5	API wiring + task controls	stale setConfiguration doc fix (packages/types/src/api.ts:116-119 → writes target the sidebar/extension-host view); `preserveOpenTabs` @PARAM JSDoc (api.ts:20-32); spec split: api-task-control.spec.ts (L96-263 + L265-305), api-set-configuration.spec.ts (happy path), api-configuration.spec.ts (getConfiguration flatten + strip-secrets, L70-102 new); PR body documents the sidebar-only API limitation (A4 major, parked)	557 measured (549+/8−, `b5875118e` re-stacked on F2 head `aa0f3b1b7`; soft-over — the Zoo-Code-Org#982 wiring-exists verdict is NOT HELD at the F1a head, so the CS `src/extension/api.ts` port is included; rationale in the PR body)	F1a, F2
F6	e2e suite (test-only)	port/extend apps/vscode-e2e/src/suite/view-state.test.ts for the vps2 split; A5 minor 2: one webview-reload/rehydration e2e test	measured at cut	F1a–F1c, F3
F7	Zoo-Code-Org#915 skeleton	skeleton port (~20 src + ~80 test)	~100 est	F6

Item	Title	Covered by
fork PR #34	Merge Zoo-Code-Org#977 (durable per-view state base) into the file-write-safety trial branch	F1 + F2 (the Zoo-Code-Org#977 content, re-cut as independent fixes)
upstream Zoo-Code-Org#967	[BUG] Per-view state lost on extension reload — no durable persistence	F1 + F2
upstream Zoo-Code-Org#984	[BUG] Add durable per-view state persistence for parallel tabs	F1 + F2
upstream Zoo-Code-Org#1065	Preserve durable editor view state across provider disposal	F2 (50-entry-cap regression; dispose-retention already satisfied in CS) — close permission-blocked (403); pointer comment posted
upstream Zoo-Code-Org#978	[BUG] Isolate parallel mode and provider profile writes	F3
upstream Zoo-Code-Org#980	[BUG] Invalidate per-view state after reset and settings import	F4
upstream Zoo-Code-Org#982	[BUG] Sync API setConfiguration through provider state wiring	F5 (sidebar-only limitation documented)
upstream Zoo-Code-Org#913	[UX] Multi-tab mode isolation — follow-up issues from PR Zoo-Code-Org#909	the series as a whole (F0–F6); individual findings dispositioned in the ledger above
upstream Zoo-Code-Org#915	[UX] Flash of incorrect mode/apiConfig when opening new editor tab	F7 (skeleton port; un-parked 2026-09-06 — wave 6)
upstream Zoo-Code-Org#929	[909] CodeRabbit Review Findings — View State Isolation Bugs	A5 re-review: 5/5 findings verified stale or superseded
upstream Zoo-Code-Org#916	[UX] Add visual indicator for per-view mode isolation	out of vps2 scope (UX polish; no fix unit)
upstream Zoo-Code-Org#917	[UX] Add save indicator for per-view API configuration changes	out of vps2 scope (UX polish; no fix unit)
upstream Zoo-Code-Org#918	[UX] Wire SettingsView Save button to syncViewStateToGlobal	out of vps2 scope (SettingsView cachedState binding already satisfied per A5; not part of CS content)
upstream Zoo-Code-Org#919	[UX] Add telemetry/tooltip for view-local state persistence limitation	out of vps2 scope (UX polish; no fix unit)
upstream Zoo-Code-Org#920	[BUG] Concurrent task history updates from parallel tabs may cause lost entries	out of vps2 scope (task history is outside the per-view state surface)
upstream Zoo-Code-Org#922	[BUG] Switching mode/apiConfig while parallel tools are running may corrupt task state	F3 (per-view mode/profile write isolation; unknown-mode hardening verified in the gap review)
upstream Zoo-Code-Org#923	[ENHANCEMENT] Support Fork-Join parallel modes with per-view state isolation	out of vps2 scope (per-view isolation layer = F1–F6; Fork-Join orchestration is a separate feature)
upstream Zoo-Code-Org#924	[ENHANCEMENT] Add "New Tab" button beside "Start New Task" for parallel mode discoverability	out of vps2 scope (UX discoverability polish; no fix unit) — same disposition as Zoo-Code-Org#916/Zoo-Code-Org#917/Zoo-Code-Org#919
upstream Zoo-Code-Org#921	[BUG] Parent-child task delegation across parallel tabs may lose state	out of vps2 scope (task delegation is outside the per-view state surface)
upstream PR Zoo-Code-Org#966	fix(webview): isolate parallel provider view state	F1/F2 content (porting source, stale head `beaf30b90`)
upstream PR Zoo-Code-Org#928	fix(webview): add loading skeleton for view state initialization	F7 (Zoo-Code-Org#915 skeleton port; un-parked 2026-09-06; its vscode.ts + webviewDidLaunch hunks redundant in CS)
upstream PR Zoo-Code-Org#979	fix(webview): isolate parallel mode and provider profile writes	F3 content (porting source, stale head `e21ff41e9`)
upstream PR Zoo-Code-Org#981	fix(webview): invalidate per-view state after reset and import	F4 content (porting source, stale head `e7512a08b`)
upstream PR Zoo-Code-Org#983	fix(api): sync setConfiguration with provider state	F5 content (porting source, stale head `2d162d6dd`)

VPS2 durable per-view state - independent-fix series (supersedes #34 + the 21-item upstream draft series) #41

Description

Unit status (maintained as PRs open/merge)

Severity rollup (six-track DTE gap review, 2026-09-05)

Phase 0 audit ledger (A5 re-review + A6 button audit)

Risk flags (quick reference)

VPS2 Recreation Plan — durable per-view state as independent fixes

Independence guarantee

Fix table

Per-fix workflow (identical for every F)

Execution waves (max 3 parallel workers)

PR body template (per fix PR)

CS hunks intentionally NOT ported (verified 2026-09-05; observed by every vps2 PR)

Parked-items register (12, all bounded)

Superseded-items disposition (24 items; 23 closed 2026-09-05, Zoo-Code-Org#1065 close permission-blocked)

Bookkeeping checklist

Non-goals

Activity

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions