[Cuarroc]

Die Live-Profilwahl bleibt in den geprüften Fenstergrößen sichtbar und verdeckt keine Tabs oder Schließen-Schaltflächen. Die Bedienung des Menüs verändert die laufende Terminalgröße nicht, und ein neuer Test schützt ausstehende Worker-Aufrufe nach einem fehlgeschlagenen Start. Die Lieferung steht als Entwurf für zwei unabhängige Reviews und die Abnahme durch Root bereit.

Report

PR: #1917 — OPEN, DRAFT, exact HEAD verified with gh pr view 1917 --json url,isDraft,state,headRefOid,baseRefName,headRefName,labels (exit 0); no labels. gh pr create --draft exited 0.

Package / Plan-ID: FU-KR08C-LIVE-PLACEMENT. Operator: UMSETZEN. Owner: srv-fu-kr08c-live-placement. Ticket: a476a3ad-787a-4b4f-a66f-9b71bf694c25. Parent 366c57e2-488c-48c1-9405-90b0cf28fab6 remains open.

• BASE: 5c15cddeccf6e0ab8333f2c72326cc94223c263e.
• HEAD / verified remote: 384eb4edb94a0705c67729680b1073cee7430830.
• Branch: codex/fu-kr08c-live-placement.
• Actual author runtime: OpenAI gpt-6-astra, effort high, observed in this worker's native header and turn_context at 2026-10-26T00:44:59.776Z.
• RAW: 245 = 243 added + 2 deleted, including tests and the single PLAN row; measured with git diff --numstat 5c15cdde...HEAD. Initial forecast was approximately 280; no size exception.
• Main rechecked before push: same BASE, CI run 40215133788 completed successfully; no open PR or duplicate branch PR at that check.
• Admission file hash matched the supplied SHA and was within its start window. Own start-check exited 0. Native goal get/create/get succeeded; worker delivery is bounded to this draft handover. The local lage-jetzt mirror was unavailable; live GitHub, git, job and process evidence was used. Only this identified author CLI was observed during the heavy-phase guards, with an exclusive existing build target, load below 8 and available memory above 3500 MiB.

Change and boundaries. TerminalLauncher.tsx:35 measures the complete action strip when the live chooser opens, chooses available space below or above that strip, clamps horizontal placement and height, and tracks strip/viewport movement until closing. Four scoped CSS lines apply the fixed overlay only inside live actions. The empty launcher retains its existing placement. TerminalPlacement.browser.mjs:10 is a committed Node test using installed Vite and Chromium, real public UI actions, and the existing rich Tauri boundary fixture; its finite server and browser close in finally. TerminalRoute.test.tsx:962 adds C1 through the real provider and a visible public openAdHoc consumer. docs/PLAN.md:260 is the only documentation change. Shared Popover/useFlip, provider, reducer, App, mock, IPC and native sources are unchanged. No dependency or license additions.

RED versus fault probe. Before production edits, the new geometry test executed successfully as a browser fixture against exact BASE production and failed on actual assertions: right edge 1123.28125 > 1100, and, in the separately ordered short case, panel/tab intersection. These were assertion failures, not imports, browser setup or missing-test failures. The final test was also replayed with the two changed production files restored temporarily from BASE, then restored byte-for-byte to the candidate: it failed again at the right edge. BASE images below bind to those BASE production files; HEAD images were captured by the canonical test runs on final HEAD. The canonical red-first.sh separately reports the new file as missing on merge-base and green on HEAD; that missing-file result is not presented as the observed geometry RED.

C1 is an already-correct BASE behavior, not a newly discovered runtime bug. Its normal run exited 0. An uncommitted, backed-up scratch fault removed only if (!launchPending) launchSessions.current = null;; the same test exited 1 because #/leitstand/terminal?worker=two became #/leitstand/terminal. The source was restored unconditionally and matches its backup SHA256 7fbfa28434ead8a0f650729720462f6eee16f2c2902783fc7536996e9c2534c0. Deferred rejection and lookup requests settle in unconditional cleanup, retained error acknowledgement is public, and original/cleanup diagnostics use the existing aggregate-error helper. No Test-First claim is made for C1; commit trailers name only the geometry test.

Commands and observed raw exits. All commands ran in the assigned worktree. Full prepush and the normal push hook identify final HEAD 384eb4ed; the tree is clean.

Command / condition	Exit	Evidence
bash scripts/sync.sh start	0	Own clean branch at BASE
npm run dev:start-check	0	13.5 GiB free, zero Cargo builds; native supplied quota file checked separately
node --test src/screens/terminal/TerminalPlacement.browser.mjs with BASE production	1	Actual overflow and wrapped-tab collision assertions, separate case orderings
npx vitest run src/screens/terminal/TerminalRoute.test.tsx -t 'disarms a settled launch'	0 / 1	BASE behavior green / narrow scratch fault red; other tests filtered, not weakened
npx vitest run src/screens/terminal/TerminalRoute.test.tsx src/sessions/SessionProvider.test.tsx src/sessions/sessionReducer.test.ts	0	68 passed: existing 67 plus C1
node --test src/screens/terminal/TerminalPlacement.browser.mjs with candidate	0	Geometry, dismissal, focus, actual clicks, close semantics, separate alert-row measurement
npm run lint	0	Explicit Node imports and browser-global declarations in the mixed-realm test
bash scripts/ci/secret-scan.sh and normal commit hooks, each commit	0	Three commits; no leaks; hook typecheck passes
bash scripts/ci/gates.sh lane prepush, initial candidate	1	ESLint did not know the new .mjs execution globals; corrected in 384eb4ed, without changing lint configuration or assertions
bash scripts/ci/gates.sh lane prepush, final HEAD	0	282 s; 235 frontend files / 1754 tests passed; 2327 Rust tests passed, 7 existing skipped; HQ/Denkraum, typecheck, lint, clippy and ordinary gates green
PLACEMENT_CAPTURE=.pa/placement-head bash scripts/ci/red-first.sh, final HEAD	0	Final named browser test green for all three exact commit trailers; screenshots captured on final HEAD
git push -u origin codex/fu-kr08c-live-placement	0	Normal full prepush hook also green, 180 s
git ls-remote origin codex/fu-kr08c-live-placement	0	Exactly 384eb4edb94a0705c67729680b1073cee7430830

Browser observations. Actual viewports are 1100×720, 760×520 and 760×260. Each capture has six tabs, all labelled Claude, selected session ses-adhoc-6, route #/leitstand/terminal, theme system/light, variant glas, reduced motion enabled. Setup uses six real public launches and waits for the selected xterm viewer/fonts and scheduled fits before taking baselines. The boundary observer records all resize_pty calls and arguments and forwards the original invocation/result unchanged. The visible terminal error transcript is existing fixture content, not a failing test run.

Coordinates below are (x, y, width, height) in CSS pixels. Full rects, labels and invocation arrays are retained in .pa/placement-base-final.log and .pa/placement-red-first-final.log; the initial 760×260 BASE collision is also recorded in the worker tool transcript and inspected image.

Viewport	BASE panel	HEAD panel	Strip, unchanged by chooser	Viewer, unchanged by chooser
1100×720	(803.28125,157.796875,320,280); right 1123.28125	(600.28125,160.796875,320,280); right 920.28125	(248.28125,110.796875,839.71875,42)	(248.28125,164.796875,839.71875,543.203125)
760×520	(348.859375,208.796875,320,280)	(145.859375,208.796875,320,280)	(231.859375,110.796875,516.140625,90)	(231.859375,212.796875,516.140625,295.203125)
760×260	(348.859375,0.796875,320,156); genuinely flipped above trigger into tabs	(145.859375,8,320,94.796875); above whole strip, bottom 102.796875	(231.859375,110.796875,516.140625,90)	(231.859375,212.796875,516.140625,35.203125)

At 1100 width, tab rectangles start at x=253.28125 + 70*i for i=0…5, y=115.796875, size 66×32; close is (686.28125,115.796875,105,32). At 760 width, tabs start at x=236.859375 + 70*i, same y/size; close wraps to (231.859375,166.796875,105,32). Every measured HEAD panel is disjoint from every tab/close rectangle. The 760×260 panel scrolls internally because only 94.796875 pixels fit above the strip; actual cancel and Escape interactions still pass. No claim extends to all possible heights, zoom factors or browser engines.

Observed total resize counts at HEAD: 760×260 13→13, 760×520 14→14, 1100×720 15→15 for opening, and unchanged after each cancel, Escape and outside dismissal. The first 12 calls belong to the six setup viewer mounts; viewport changes are outside chooser-only baselines. Strip/viewer rects are deeply equal before/open/after each dismissal. Cancel and Escape return focus to the public trigger. Actual pointer clicks with the chooser open select the second tab and close it; the sole kill is exactly {sessionId:"ses-adhoc-2"}. Attaching then closing worker w1 while the chooser is open leaves that kill list unchanged. Deliberate selection/remount is measured separately: 16→18 calls. A fresh launchFailure=1 document attaches a worker before its first spawn; error opening leaves count 0→0 and the viewer unchanged, while closing exposes the accepted alert row and produces exactly 0→1, strip height 42→73.5 and viewer height 543.203125→511.703125. Reopening publicly acknowledges the error.

Inspected images. All six images were inspected after final capture; full panels and controls were checked. Artifacts are untracked evidence, not substitutes for the committed browser regression. SHA256:

Artifact	SHA256
.pa/placement-base/1100x720.png	0e35543efa4ff8fa10d43c1d382a0e036f82046250252e6965837fc181c08b8c
.pa/placement-base/760x520.png	932f8706dc611dffa861dc166ab28c436e6e6cad3ea8f07ad8968ff81ed4dc5e
.pa/placement-base/760x260.png	696882cfcf96d27a55286a728a99246c10708955b357bbc65b3f8520edcfea3e
.pa/placement-head/1100x720.png	a39f67014c0da399be4a2ec778c7974c607ec14b5fd983dd897d0f3d4e97bdc7
.pa/placement-head/760x520.png	ddde2782c6492b0eb717f258d794a4d746262012ad2b69fc883d86708bafd5c5
.pa/placement-head/760x260.png	03f01c6602600a66c219b1b7eaed8edf528eb9370313799d4b87d6f25322844f

P1 — viewport fit

With at least six real terminal tabs in the recorded narrow viewport, the live chooser's full panel stays inside the viewport at the right edge, preserves the existing enabled-profile and truthful-error behavior, and is proved by actual browser rectangles and inspected BASE/HEAD captures rather than an alignment attribute alone.

Disposition: evidenced in the table, inspected captures, real browser test and preserved route suite. BASE right-edge overflow is red; HEAD is inside. Enabled-profile filtering remains covered by the literal-payload public route tests; the fresh browser failure remains truthful.

P2 — wrapped controls and flip

In the recorded short or wrapped live-action layout, the open chooser never intersects the tab or close-control rectangles, including an upward placement when it is reached by real layout, while existing shared Popover and provider code remain unchanged and every unproven placement is stated for Root before acceptance.

Disposition: evidenced at 760×520 and 760×260. The latter reaches upward placement through real geometry on both BASE and HEAD; no side flag or geometry is faked. HEAD positions above the entire strip. Shared code is unchanged. Other viewport/zoom/engine combinations remain unproven; the extremely short case uses internal panel scrolling.

P3 — zero chooser resize and usable clicks

Opening, cancelling, escaping or outside-dismissing the chooser leaves the strip and viewer geometry unchanged and adds no resize_pty call after the viewer settles, while actual pointer clicks on another tab and Tab schließen with the chooser still open take effect with the existing worker versus ad-hoc close semantics; alert-row and deliberate viewer-remount resizes are measured separately.

Disposition: evidenced by all-call snapshots, deep rect equality, real Playwright pointer actions and literal kill arguments. Separate remount and alert-row counts are above. Focus measurements cover the public cancel/Escape trigger behavior only.

P4 — public no-insertion disarm regression

After an initiated request rejects without insertion and publicly settles, an unrelated session insertion while the route remains mounted does not consume a stale launch snapshot to clear the current worker deep link or cancel its pending lookup; the regression uses the real public provider and route boundary, detects a scratch disarm fault, and adds no private reset, provider mock or production state change.

Disposition: evidenced by TerminalRoute.test.tsx:962, the normal green test and scratch-fault red result. The mounted route preserves worker=two through unrelated insertion and displays its eventual explicit result. No route implementation change remains.

P5 — real RED and preserved lifecycle

The geometry regression compiles and fails against BASE before the minimal placement fix, passes on the exact candidate through the real browser boundary, and retains the existing pending/remount/error and explicit-restart regression contracts with unconditional public request cleanup and no weakened assertion, skip, extra retry or timeout change.

Disposition: evidenced by the manual actual-geometry RED, canonical final-head green, 68 targeted scenarios and full frontend suite. C4 explicit restart, shared pending and retained error scenarios remain unchanged. The 350 ms browser quiescence wait allows scheduled fits before measurements; no test timeout/retry/skip setting was changed. The initial lint failure was corrected by declarations/imports only.

P6 — bounded reviewed delivery

The new follow-up stays within 300 RAW lines, terminal implementation/test files and its single mandatory own PLAN row, passes targeted tests and full own prepush with normal verified push, and returns one DRAFT for Root's two non-author-family reviews; RESTORE, split, D1, exited-tab eligibility, native, installed/effective, whole-Glass, UAT and parent acceptance remain open.

Disposition: implementation and evidence delivered at 245 RAW; normal push verified; this PR is a DRAFT. Root owns reviews, acceptance and Mergify. No reviewer/team/QA worker was started here. Parent and the listed broader acceptance duties remain open; predecessor #1915 is not reopened.

Architecture drift: gate passes; no allowlist changes or new native/API method.

Review: abgeschlossen (Stufe B), drei Fremdfamilien-Reviews am Kopf 384eb4ed; Disposition und Root-Abnahme unten.

NICHT ABGEDECKT

• Linux does not compile #[cfg(windows)] tests, including run_in_pty, npm_shim, ConPTY and DPAPI, or Windows clippy arms. The Windows lane runs in the merge queue.
• Full prepush reports 7 existing Rust skips; HQ/Denkraum also retain their existing platform/fixture skips (3 and 1). This package adds no skips. The 1754 frontend tests and 68 targeted scenarios all passed.
• The generic browser smoke suite, frontend production build and full workflow gates are outside the local prepush lane. The separately executed committed Chromium regression covers only the scenarios specified above.
• External services, updater/OmniRoute/mirror state, native/installed/effective delivery, assistive-technology behavior, other browser engines and unrecorded layouts are not proved.
• RESTORE, split, broader D1, exited-tab eligibility, whole-Glass/UAT, two independent reviews, Root/parent acceptance and Mergify remain open. The default empty chooser's broader geometry is unchanged and not newly certified.

Next owner: Root, through the existing checked Stabs handover chain, for two family reviews and acceptance. Local report and evidence are retained; no history rewrite, new installation, release or service change was performed.

Root review disposition and acceptance (2026-10-26, head 384eb4edb94a0705c67729680b1073cee7430830)

Tier B (TS/CSS, tests, own PLAN row; no seam). Author family OpenAI (Codex gpt-6-astra). Reviews bound to head 384eb4e:

#	Reviewer (observed identity)	Verdict
R1	Anthropic, Claude Opus 5.5 claude-opus-5-5[1m] (PC subagent, read-only)	APPROVE WITH CONDITIONS (A1 disclosure, capture hashes)
R2	Moonshot, Kimi CLI kimi-code/k3 (PC, wire.jsonl of session session_8dc2545a-24d2-495f-9101-a1c19e8066bc)	APPROVE
R3	Anthropic, Claude Sonnet 5.5 claude-sonnet-5-5 (server job srv-review-pr1917-claude-0216, report ~/pa-orch/reports/review-pr1917-claude.md)	APPROVE WITH CONDITIONS (P1 PLAN row, A1)

No high finding. Disposition (acceptance lines quoted from the brief before any follow-up):

ID	Severity	Finding	Disposition
R1917-A1 (R1, R3)	medium	TerminalPlacement.browser.mjs runs in no gate lane after merge (gates.sh, vitest.config.ts:26, playwright.config.ts:4)	accepted as disclosure: NICHT ABGEDECKT line added below by Root; code change is out of this brief (gate files not in FILES). Follow-up package FU-1917-GATE-ALIGN: wire into linux_2 next to hq-visual (scripts/ci/gates.sh:197, package.json).
R1917-P1 (R3)	medium	PLAN row 259 (KR-08c-2b-LIVE) still says "Delta-Review/Root-Abnahme offen", #1915 merged 5c15cdde	follow-up: P6 scope is "its single mandatory own PLAN row"; row 259 belongs to the predecessor. Fix goes to the running PLAN-PFLEGEN package (P1, job 184) or FU-1917-GATE-ALIGN; Root accepted #1915 on 25.10.
R1917-A2 (R1) = R1917-D1 (R3)	low	panel always right-aligned to the trigger; at 760 px it reaches ~86 px over the sidebar (760×520) and the header (760×260)	follow-up in FU-1917-GATE-ALIGN: start-align + clamp (TerminalLauncher.tsx:49), add header/sidebar rects to the test. Acceptance P1 "inside the viewport at the right edge" and P2 "clear of wrapped tabs and close" hold; Root inspected the captures (below).
R1917-B1 (R1)	low	clamp uses window.innerWidth, which includes a classic scrollbar	follow-up (same package): documentElement.clientWidth
R1917-A1 (R2)	low	position: fixed depends on no ancestor containing block (transform/filter/backdrop-filter)	recorded; R1 checked glass.css on the measured path, none today. Follow-up note in the same package (comment + test).
R1917-B1 (R2)	low	strip/viewport tracking while the chooser is open is not exercised	follow-up (same package): resize with chooser open
R1917-T1 (R3), R1917-C1 (R2)	low/nit	one environment (light, glas, reduced motion), fixed 350 ms settle	NICHT ABGEDECKT line added below
R1917-S1 (R3), R1917-B3 (R1), R1917-E1 (R2)	low/nit	second placement mechanism beside the design-system Popover, selector specificity, silent degradation	recorded as design follow-up (Popover opt-in); not scheduled
R1917-B2, C1, C2, C3 (R1), R1917-D1 (R2)	nit	anchor only from onClick, scroll re-measure cost, cleanup rejection masking, near-zero height	recorded, no action

NICHT ABGEDECKT (added by Root after review):

• No gate lane, npm script or workflow runs src/screens/terminal/TerminalPlacement.browser.mjs after merge; it ran only in this PR's red-first job and in the reviewers' runs (R1917-A1). Follow-up FU-1917-GATE-ALIGN.
• The browser regression pins light theme, variant glas and reduced motion and settles with a fixed 350 ms wait; animated closing, dark mode and the variants klar/abend/nebel are not exercised (R1917-T1).
• Header, project picker, search and sidebar controls are not in the test's disjointness set; at 760 px width the open panel covers part of them until it is dismissed (R1917-A2/D1).

Root checks: the six capture hashes in the table above match sha256sum of fu182-browser-evidence.tgz on the server and locally (2026-10-26 01:25Z). Root inspected the three HEAD captures: 1100×720 panel below the strip, right edge inside the viewport; 760×520 panel below the wrapped strip, clear of tabs and close, overlapping the sidebar edge; 760×260 panel flipped above the strip at 95 px height, scrolling internally. Codex connector comments on this head: none. PLAN row 260 present. Acceptance: P1–P6 met at head 384eb4e; integration continues with ready, CI on head, review-ok bound to 384eb4e and @Mergifyio queue.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-26T01:30:03.272947Z	384eb4e	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 384eb4edb9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Aktualisiere die eigene Planzeile auf den abgenommenen Stand

Die neue Paketzeile bezeichnet zwei unabhängige Reviews und die Root-Abnahme weiterhin als „ausstehend“, obwohl der Bericht dieses Commits unter „Root review disposition and acceptance“ drei abgeschlossene Fremdfamilien-Reviews und die Annahme von P1–P6 am Kopf 384eb4ed dokumentiert. Damit hinterlässt der Merge im einzigen verbindlichen Plan bereits erledigte Prüfschritte als offene Restpflichten. Bitte übernimm die belegten Review- und Abnahmeergebnisse in diese eigene Paketzeile; die ausdrücklich offenen Eltern-, Native- und UAT-Abnahmen bleiben davon unberührt. Der vorgeschriebene Plan-ID-Abgleich gilt auch für diese Zeile.

AGENTS.md reference: AGENTS.md:L74-L76

Useful? React with 👍 / 👎.

[Cuarroc]

Root acceptance (INTEGRIEREN), head 384eb4edb94a0705c67729680b1073cee7430830, 2026-10-26T01:49:51Z: CI run 40254281274 green on this head (gates linux/windows, red-first, linux_1/linux_2, windows_1/2/native). Reviews bound to this head: Opus 5.5 (Anthropic) APPROVE WITH CONDITIONS, Kimi kimi-code/k3 (Moonshot) APPROVE, Sonnet 5.5 server job (Anthropic) APPROVE WITH CONDITIONS; conditions evidenced in the PR body (NICHT ABGEDECKT addendum, capture hashes verified, follow-ups FU-1917-GATE-ALIGN / PLAN row 259 via #1918). Codex connector P2 on docs/PLAN.md:260 (own row still says reviews/acceptance pending): author-family advisory; the row states delivery-time status, acceptance line P6 is not breached; the post-merge wording goes into the #1918 PLAN fix (same NACHBESSERN job) instead of a new push here. Setting review-ok bound to this SHA; any later push voids it.

[Cuarroc]

@Mergifyio queue

[mergify]

Merge Queue Status

• ✅ Entered queue — 2026-10-26 01:50 UTC · Rule: default · triggered by @Cuarroc with the @mergifyio queue command
• ✅ Checks passed · on draft merge queue: checking #1917 on main (5c15cdd) #1920
• ✅ Merged — 2026-10-26 02:18 UTC · at be8b70356ac888bfafdad6a7250a9155cdd77063 · merge

This pull request spent 27 minutes 38 seconds in the queue, including 26 minutes 59 seconds running CI.

Required conditions to merge

• check-success = gates (linux)
• check-success = gates (windows)
• check-success = red-first
• github-review-approved [🛡 GitHub repository ruleset rule main-pr-required-checks]
  • fix: keep live terminal chooser inside viewport and clear of tabs #1917
• label != do-not-merge
  • fix: keep live terminal chooser inside viewport and clear of tabs #1917
• any of [🛡 GitHub repository ruleset rule main-pr-required-checks]:
  • check-success = @github-actions/gates (linux)
  • check-neutral = @github-actions/gates (linux)
  • check-skipped = @github-actions/gates (linux)
• any of [🛡 GitHub repository ruleset rule main-pr-required-checks]:
  • check-success = @github-actions/gates (windows)
  • check-neutral = @github-actions/gates (windows)
  • check-skipped = @github-actions/gates (windows)
• any of [🛡 GitHub repository ruleset rule main-pr-required-checks]:
  • check-success = @github-actions/red-first
  • check-neutral = @github-actions/red-first
  • check-skipped = @github-actions/red-first
• any of [🛡 GitHub repository ruleset rule main-pr-required-checks]:
  • check-neutral = @mergify/Mergify Merge Protections
  • check-skipped = @mergify/Mergify Merge Protections
  • check-success = @mergify/Mergify Merge Protections