[brennanb2025]

	Files	Added	Deleted	Net
Test	17	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​1875	$\color{#cf222e}{\Huge{\mathbf{−}}}$​133	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​1742
Prod	43	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​1223	$\color{#cf222e}{\Huge{\mathbf{−}}}$​1205	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​18

The problem

Orca starts a terminal in two ways. The app window starts one when a terminal pane opens. Everything else (the orca command line, the phone app, agents and automations) asks Orca to start one directly. Each way had its own copy of the start-up code, and the two copies made their own decisions.

When both tried to start the same terminal at the same moment, whichever got there first decided what the other one saw. In practice:

• If the program in the terminal quit immediately, the window and the command line could disagree about what happened. One might show nothing while the other reported success.
• A start that was cancelled part-way could leave things reserved that nothing would ever release.
• Fixes had to be made twice, and the two copies kept drifting apart.

What this changes

There is now a single start-up path for every terminal, whoever asks for it.

• If two requests start the same terminal together, the second one joins the first. Only one program is started, and both callers get the same result.
• If the program quits during start-up, the window still shows its last output, so you can read why it stopped. The command line, phone, agents and automations still get a clear start-up failure, as before.
• A new terminal starts at the size that was asked for. Reconnecting to a terminal that is already running keeps its current size.
• A cancelled or refused start now releases everything it reserved.
• Main's recent change (A chat never starts writing a conversation a terminal is still writing, and restored panes respect open chats #30930) that stops a chat and a terminal from writing to the same saved conversation now runs inside this single path. The check happens once, and every caller that joined the start sees the same answer.

What stays the same

There are no new settings, no change to what gets saved, and no change to what Orca sends between machines. Older paired servers and older phone apps keep working as they do today. Remote (SSH) and multi-machine setups still start terminals on the machine that owns them. Commands people type, and the commands agents are launched with, are unchanged.

Linked issue

Tracked in https://linear.app/stably/issue/STA-11756.

How it was tested

• An independent review ran three rounds over the whole change, compared it with the common pattern for this kind of code, and checked how it scales with many terminals.
• Automated tests cover two starts racing each other, a program quitting during start-up, cancelled starts, terminal sizes, cleanup, and the phone, remote and automation callers.
• Real-app testing on macOS, native Windows and Windows with WSL covered:
  • terminals started from the window and from the command line;
  • restarting Orca with terminals open, including upgrading from the published app with saved terminals;
  • hiding, showing and reloading panes;
  • cancelling starts and programs that quit at once;
  • Claude, Codex and OpenCode sessions continuing to work.
• Since that testing, this branch has only merged in main. Main merges don't need retesting under our rules: CI checks the merged result.

Current CI and review status, with every screenshot, is in the review status comment.

Existing problems this PR does not fix

The same testing turned up four problems that already exist in the published app and that this PR doesn't change:

• On WSL, a plain terminal forgets the folder you chose after Orca restarts. Tracked in STA-12099. The published app's screenshot of this timed out, so it is UNMEASURED; the saved data and the folder the terminal actually opened in show the behaviour.
• A very short-lived terminal started from outside the window (for example by the command line) can lose its last line of output. Tracked in STA-12101.
• On native Windows, after Restart, terminal rows can show an extra folder suffix. Tracked in STA-12029. This PR · published app.
• Reconnecting to a terminal can repeat a little output when the saved picture of the screen already contains it. This is temporary, tracked in STA-11855. This PR keeps the existing protection against losing newer output.

One observation stayed unexplained: Once, on native Windows after Restart, two whole-window screenshots showed rows from the previous terminal; it did not recur in a paired rerun on the same build and on the previous build, and its cause is unknown.

Differences from the common pattern

• Intended: Orca still chooses which machine runs each terminal, still refuses to reuse a terminal that has been replaced, and still supports older servers and phones. This product needs all three: it runs terminals on several machines, and clients and servers update at different times.
• Intended: a window shows the final output when a start fails, while other callers get an error. A window has a screen to show output on; the others don't.
• Temporary: how hidden terminals receive output still depends on who started them. Follow-up: STA-12188.
• Temporary: the window still prepares part of its launch settings itself. Follow-up: STA-11757.
• Temporary: each caller still works out where to run a terminal in its own way. Being unified in fix(terminal): match Windows workspace runtime across launch paths #31980.

Agent skill upstream boundary

• This change copies or mechanically translates no upstream skill-installer material.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

• Last reviewed commit: 0f9abfc
• Latest commit, not reviewed: fa3a98d

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

No actionable comments were generated in the recent review.

ℹ️ Recent review info

⚙️ Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: c759490b-c8ee-4e80-9045-366ddef555bd

📥 Commits

Reviewing files that changed from the base of the PR and between 62c15b5 and 0f9abfc.

📒 Files selected for processing: 6

• src/main/ipc/pty-launch-race-parity.test.ts
• src/main/ipc/pty-pane-materialization-race.test.ts
• src/main/ipc/pty/spawn-outcome.ts
• src/main/runtime/cli-terminal-create-host-session-binding.test.ts
• src/renderer/src/components/terminal-pane/pty-transport-connect-spawn.test.ts
• src/shared/terminal-attach-result.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 3 remain after this review.

Walkthrough

The change adds shared orchestration for IPC and runtime PTY spawns, with committed, adopted, and registration-exited outcomes. Pane reservations retain spawn operations and outcomes for adoption and registration-exit handling. IPC and runtime entry points use the shared lifecycle and new-incarnation ownership state. Terminal creation tracks agent-team launch allocations through spawn outcomes. Tests cover spawn races, pane restoration, failure cleanup, and outcome handling.

Priority: Normal

Merge Risk: Minimal · up to 0f9ab

The previously reported lint blocker is fixed. No remaining issue identified in these changes prevents merging after normal checks.

Pre-merge checks: 4 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 33.93% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 56 functions across 46 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

Passed checks: 4 passed

Check name	Status	Explanation
Title check		The title clearly summarizes the primary change: consolidating terminal starts under one host operation.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.
Description check		The description is detailed and covers the problem, changes, rationale, linked issue, testing, compatibility, known limitations, and upstream boundary. It does not include every template heading or ch…

Full details: Description check

Explanation

The description is detailed and covers the problem, changes, rationale, linked issue, testing, compatibility, known limitations, and upstream boundary. It does not include every template heading or checklist item, but the required information is mostly present, including references to review screenshots and outstanding validation.

✨ Finishing Touches: 💡 2

📝 Generate docstrings: 💡

• Commit to this branch
• Create a new PR

🛠️ Fix failing CI checks: 💡

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 6ba21a05-cb6b-4796-b77c-30c3493efc45

📥 Commits

Reviewing files that changed from the base of the PR and between a45bd06 and d1156f5.

📒 Files selected for processing: 44

• src/main/ipc/pty-hidden-at-spawn-mark.test.ts
• src/main/ipc/pty-launch-race-parity.test.ts
• src/main/ipc/pty-pane-materialization-race.test.ts
• src/main/ipc/pty/ipc/spawn-begin.ts
• src/main/ipc/pty/ipc/spawn-commit-run-facts.test.ts
• src/main/ipc/pty/ipc/spawn-commit.ts
• src/main/ipc/pty/ipc/spawn-env-codex.ts
• src/main/ipc/pty/ipc/spawn-execute.ts
• src/main/ipc/pty/ipc/spawn-options.ts
• src/main/ipc/pty/ipc/spawn-preflight.ts
• src/main/ipc/pty/ipc/spawn-reattach-size-cache.test.ts
• src/main/ipc/pty/ipc/spawn-run.ts
• src/main/ipc/pty/ipc/spawn-state.ts
• src/main/ipc/pty/ipc/spawn-types.ts
• src/main/ipc/pty/pane/adopt-stable.ts
• src/main/ipc/pty/pane/complete-pane-adoption.ts
• src/main/ipc/pty/pane/pane-spawn-operation.ts
• src/main/ipc/pty/pane/spawn-reservation.ts
• src/main/ipc/pty/runtime/spawn-commit-response.ts
• src/main/ipc/pty/runtime/spawn-commit.ts
• src/main/ipc/pty/runtime/spawn-early.ts
• src/main/ipc/pty/runtime/spawn-hidden-delivery.ts
• src/main/ipc/pty/runtime/spawn-options.ts
• src/main/ipc/pty/runtime/spawn-preflight.ts
• src/main/ipc/pty/runtime/spawn-state.ts
• src/main/ipc/pty/runtime/spawn.ts
• src/main/ipc/pty/spawn-begin.ts
• src/main/ipc/pty/spawn-commit-assembly.ts
• src/main/ipc/pty/spawn-commit-publication.ts
• src/main/ipc/pty/spawn-execute-failure.ts
• src/main/ipc/pty/spawn-execute.ts
• src/main/ipc/pty/spawn-operation-state.ts
• src/main/ipc/pty/spawn-operation.ts
• src/main/ipc/pty/spawn-options-assembly.ts
• src/main/ipc/pty/spawn-outcome.ts
• src/main/opencode/opencode-startup-prompt-owner.ts
• src/main/persistence/loading-store/pty-spawn-exit-durability.test.ts
• src/main/runtime/claude-agent-teams-service.ts
• src/main/runtime/cli-terminal-create-host-session-binding.test.ts
• src/main/runtime/runtime-pane-admission.ts
• src/main/runtime/runtime-pty-controller-contract.ts
• src/main/runtime/terminal/launch/terminal-create-team-allocation.ts
• src/main/runtime/terminal/launch/terminal-create.ts
• src/main/runtime/terminal/launch/terminal-split-source-authority.ts

💤 Files with no reviewable changes: 3

• src/main/ipc/pty/ipc/spawn-commit-run-facts.test.ts
• src/main/ipc/pty/runtime/spawn-commit-response.ts
• src/main/ipc/pty/ipc/spawn-execute.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 8 remain after this review.

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to GitHub limitations.

⚠️ Outside diff range comments (1)

🟡 Minor · Restore provisional PTY size on every post-preparation exit. · spawn-operation.ts:86-89

src/main/ipc/pty/spawn-operation.ts:86-89
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Restore provisional PTY size on every post-preparation exit.

After preparation succeeds, the outer error handler and the registration-exited path release preparation without restoring the provisional PTY size. Route these paths through one guarded retirement helper. Keep release-only cleanup when preparation is not active.

Suggested fix

   let prepared = false
+  const releasePrepared = (reattachId?: string) => {
+    if (prepared) {
+      retirePreparation(adapter, reattachId)
+      prepared = false
+    } else {
+      adapter.releasePreparation?.()
+    }
+  }
   const finish = (outcome: PtySpawnOutcome) => {
     ctx.paneSpawnOperation?.acquisition?.settle(outcome)
     if (outcome.kind === 'registration-exited') {
-      adapter.releasePreparation?.()
+      releasePrepared()
       adapter.releaseHiddenMark?.()
     }
@@
           .catch((error: unknown) => {
             restoreProvisionalPtySize({
               sessionId: adapter.facts().sessionId,
@@
               sessionSizeBeforeAttach: ctx.sessionSizeBeforeAttach
             })
+            prepared = false
             throw error
           })
@@
     if (prepared) {
-      retirePreparation(adapter, adopted.result.id)
+      releasePrepared(adopted.result.id)
     }
@@
     if (prepared) {
-      retirePreparation(adapter)
+      releasePrepared()
     }
@@
       restoreProvisionalPtySize({
         sessionId: adapter.facts().sessionId,
@@
         sessionSizeBeforeAttach: ctx.sessionSizeBeforeAttach
       })
+      prepared = false
       throw error
@@
-          retirePreparation(adapter, adopted.result.id)
+          releasePrepared(adopted.result.id)
@@
-    adapter.releasePreparation?.()
+    releasePrepared()

---

ℹ️ Review info

⚙️ Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 40de8e2e-4aba-42dc-8da7-d6522e9139fc

📥 Commits

Reviewing files that changed from the base of the PR and between 04b4271 and a311773.

📒 Files selected for processing: 4

• src/main/ipc/pty-agent-pane-launch.test.ts
• src/main/ipc/pty-launch-race-parity.test.ts
• src/main/ipc/pty/pane/adopt-stable.ts
• src/main/ipc/pty/spawn-operation.ts

🚧 Files skipped from review as they are similar to previous changes: 2

• src/main/ipc/pty-launch-race-parity.test.ts
• src/main/ipc/pty/pane/adopt-stable.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 5 remain after this review.

[brennanb2025]

Concurrent desktop and host terminal starts had separate lifecycle owners. A caller could settle or clean up the same pane differently from its concurrent caller. This PR gives both entries one host operation, retains final output for eligible desktop registration exits, and preserves startup refusal for host callers.

Independent review completed three correctness rounds, the final whole-mechanism comparison, scaling review and readiness checklist. Fixes cover closed-pane dispatch, canonical exited outcomes, acquired geometry rollback, exact team allocation cleanup and OpenCode prompt cancellation. No open introduced P0/P1 remains. Deterministic tests cover both winner orders, registration durability, withdrawn/successor refusal, proof-carrying snapshots and cleanup; these are distinct from the live observations below.

Main integration commits 72940396e117d07c7f56ce3d4df7676179557577, f7f4feafbd5bb2288b477ac5902f241e2360abee, 8a71108c05019ff1e8ec91e2767589dab14c5e53, be1251a66737012f8a6aa64839f6a3570b635bbe, bbea78e80f1687255af875688628567d025558c5 and fa3a98d3630f89b47774bb857a318d63d9b1ad1d integrate main through 5ca8881871788ada5b3ef2bb4a3530e0faa2423e. The latest merge includes #33715’s localization repair, which superseded the closed #33711 / #33710. It merged automatically without conflicts or manual source edits, retained main’s lockfile, and passed every task of node_modules/.bin/lint-staged --diff=MERGE_HEAD --concurrent false. The common spawn-operation files are unchanged by this merge. Accepted QA at 9393ee021de039b455773758e580e6221f9c75ee carries across the merges; no new QA run is claimed. Fresh CI passed on fa3a98d3630f89b47774bb857a318d63d9b1ad1d: all seven workflows succeeded, with 41 passing and 17 skipped checks. Static analysis and all compiler projects passed, including localization coverage. macOS Intel persistence passed in 6m30s; the previous head’s artifact-download ENOTFOUND failure supplied no test verdict and did not recur. WSL Fedora passed on attempt 2 after CI’s automatic frozen-runner recovery cancelled its first attempt at the time cap and retried that job; the other WSL jobs retained their successful results. No manual CI rerun was started. Skipped checks are not counted as passes.

The #30930 composition makes physical fresh starts decide saved-conversation admission, unwinds a refused claim before its plain shell starts, and gives joined views one completed result. Cleanup retains the existing IPC-drain versus runtime-rejection distinction. Actual-owner launch metadata and remembered restore decisions survive producer finalization. Targeted connected, race, geometry, cleanup, consumer, hidden-delivery and scaling tests passed. Narrow resolved-hunk review returned AGREE / PASS at be1251a66737012f8a6aa64839f6a3570b635bbe, with no proved composition regression or targeted QA trigger.

Test-only commit 7eec8f7a6a1fc7cb76fafccaa9b66e7f2081d6af consolidated duplicate mock imports in two test files without changing assertions or shipped code. Type-contract correction fd0fb1ad27d25ee05b294371c15ef49b4d8b706f retained the IPC boundary’s existing unknown token/origin input types in the common declaration. Existing validators remain at use; the correction contains only types and changes no executable behavior or run-row flow. Its unbiased corrected-hunk review returned AGREE; five named files / 58 tests and all four compiler projects passed before the latest main merge. Both normal pre-commit hooks passed. QA9393 carries across these corrections. The merged head is clean and conflict-free; the final readiness helper is run after this status update.

One real QA campaign, with targeted completions of unfinished flows:

Platform	Observed result
macOS	Real Claude, Codex and OpenCode useful replies; shipped 1.4.237 state through graceful upgrade; desktop/CLI starts and input/output, saved restore/reveal/reload, geometry/wrapping, hidden terminals, restart/exit, continued teams and a main-layout subset.
Native Windows	Shipped 1.4.205 state through graceful same-profile upgrade; Claude/Codex useful second/post-reload replies, ConPTY/daemon identity, CLI create/read/send/reveal, 101×31→135×60 geometry, retained exit7 output and useful Restart successor.
Protected WSL	Published 1.4.237 state through graceful upgrade; Claude/Codex useful replies and reload, guest identity, desktop/loopback CLI, hidden/reveal, cancellation/successor. Explicit 120×40 before-view observation retained the same guest/count through reveal/input (later89×44); outer exit7 kept final marker/output/code7.

Unchanged phone, automation, remote/mixed-version and multi-host consumers are covered by read/write-to-result contracts; mobile CI exercised 27 launch, 19 action and 7 history tests. No additional phone/emulator/physical SSH or all-agent campaign is claimed. Scaling tests vary callers/panes/tabs/layout from4 to40 and verify operation/pending-retention bounds; all11 assertions passed on Ubuntu/macOS/Windows on the QA head. Compiler checks and all seven CI workflows passed on the latest main merge.

Known results and limits remain explicit:

• WSL saved cwd was lost on the target and on a distinct published 1.4.237→1.4.237 control, while title/tab/leaf/layout survived. Tracked in STA-12099. The baseline restore screenshot timed out at 5s and 30s and is UNMEASURED; raw saved-state and actual guest pwd establish the stated observation. Baseline agreement neither repairs the target nor identifies a source cause.
• Short-lived host-created terminals returned success but lost their final marker on both target and baseline; WSL tracking is STA-12101. The corresponding macOS comparison also matched. Ordinary desktop exit 0 closure is distinct from retained exit 7 output. Provider byte-zero and registration timing were not measured live; exact registration/withdrawal/rollback/sequence/team race contracts use deterministic tests.
• Once, on native Windows after Restart, two whole-window screenshots showed rows from the previous terminal; it did not recur in a paired rerun on the same build and on the previous build, and its cause is unknown.
• The narrower native extra-cwd suffix matched the baseline and target and is tracked separately in STA-12029. It does not explain the preceding unknown observation. Target control: Baseline control: .
• Pre-existing image-ahead replay duplication remains temporary, tracked in STA-11855. This PR keeps the newer-output protection without implementing that separate fix. Hidden-delivery lifetimes remain temporary to STA-12188; base/assembled launch environments remain with STA-11757; runtime-resolution inputs remain owned by fix(terminal): match Windows workspace runtime across launch paths #31980. Selected-host/incarnation/mixed-version behavior and window final-output versus host-refusal projection are intended requirements.

Claude required an absolute same-session resume/manual --continue path where its configured installed shim was missing; useful replies and reload continuity were then observed. Main-layout Mac execution began after ticket 1979 had reported expiry; the later 2071 admission is not retroactive. Main-layout shell continuity was not established merely by later usability. Native Settings had DOM/pre-upgrade image evidence but no assigned post-upgrade Settings PNG; this PR changes no Settings appearance. Original native QA-profile inspection created SQLite sidecars/touched metadata; DB and 29 selected source hashes stayed unchanged, so full metadata preservation is not claimed. The native paired baseline package had stale metadata; provenance used checkout/API/rebuild/hash evidence. Exact scheduler ordering, heap/CPU/RSS/latency and every old-client/device combination are not claimed as live measurements.

Recorded test copies stopped through their owned graceful/recorded-identity paths; gate tickets, leads and collection shells are closed. WSL protected cleanup includes the launcher's guest-cleanup receipt and idle distro never-started check, not an inferred guest census. Target package/source integrity and retained evidence were checked. Real config hashes were unchanged during WSL2161; a Codex-config difference already present versus2093 and the Mac config difference are report-only observations with unknown attribution.

For Brennan’s review

The whole-window images below show the observed flows; static images do not establish scheduler ordering or source attribution. Baseline images accompany the actual paired comparisons above and below.

macOS

Evidence	Screenshot
Fresh Claude folder pwd/reply QA1979_FRESH_FOLDER_CLAUDE_OK; no visible credential.
Fresh Codex git pwd/reply QA1979_FRESH_CODEX_OK; no visible credential.
Saved Claude conversation shows old reply and both current replies through QA1979_CURRENT_CLAUDE_2; no visible credential.
Saved Codex conversation shows old reply and both current replies through QA1979_CURRENT_CODEX_2; no visible credential.
Attached host shell visibly shows QA1979_HOST_ONCE, QA1979_HOST_ATTACHED and 63x133 grid; no visible credential.
Fresh shell reports 63x133, wraps QA1979_FRESH_GRID long output and prints checkout cwd; no visible credential.
Host and split panes show CLI_SEND_OK, SPLIT_ONCE and split's earlier 24x80 checkpoint; static image does not prove scheduling/counts; no visible credential.
Old/restored shell history, QA1979_OLD_SHELL_FINAL7 and Terminal exited code 7 with output preserved/Restart control; visible CLI IDs are runtime identifiers, no visible credential.
OpenCode Kimi K3.1 Pro shows pwd result and QA1979_OPENCODE_STARTUP_OK; no visible credential.
Claude team leader pwd/reply and continued reply with nonsecret ORCA_AGENT_TEAMS_TEAM_ID; no visible credential.
Overlap pane shows QA1979_OVERLAP_ONCE, continued ATTACHED command, cwd and 63x133; exact scheduler ordering is not proved; no visible credential.
Saved Claude conversation includes QA1979_MAIN_CLAUDE_OK and QA1979_MAIN_RELOAD_OK; image alone does not establish compile mode or lease admission; no visible credential.
Baseline 92dfb4f QA1979_HOST_EXIT tab displays empty live shell prompt, without final marker or exit overlay; relevant failure comparison, no visible credential.
Target 9393ee0 short-exit host tab displays empty live shell prompt, without final marker or exit overlay; relevant failure comparison, no visible credential.

Native Windows

Evidence	Screenshot
Whole app; shipped-state Codex SAVED WIN2162-CODEX-1046 in folder workspace with model banner and history. Notices present.
Whole app Settings Terminal; PowerShell selected. Pre-upgrade image only; scrollback row is offscreen.
Whole app restored Native2162-Shell prints WIN2162-RESTORED-SHELL-1100, PID19596, checkout cwd,135,60.
Whole app Terminal3 split; separate WIN2162-NEW-LEFT-1103/PID3548 and WIN2162-NEW-RIGHT-1103/PID2220 in adjacent panes.
Whole app Native2162-Host shows WIN2162-HOST-FIRST and WIN2162-HOST-CLI-SEND-1105/PID19396.
Whole app Terminal5 displays original geometry marker101/31/PID10228 and completed AFTER-RELOAD marker135/60/samePID10228.
Whole app Native2162-Claude; historical shell parser error retained, Claude2.1.221/Opus5; SAVED, token-free RESTORED, SECOND explanation, POSTRELOAD marker and useful history-continuity explanation all visible.
Whole app folder workspace; Codex0.162.1/GPT6.1Sol high, SAVED/RESTORED/SECOND42 and completed POSTRELOAD42 plus continuity explanation all visible.
Whole app right predecessor retains WIN2162-EXIT-TAIL-RIGHT-1124/PID2220 and Terminal exited/exit code7/output preserved/Close/Restart overlay.
Whole app Terminal3 successor shows WIN2162-RESTART-SUCCESS-1125/PID14372/cwd/42 mixed with stale WIN2162-NEW-RIGHT-1103 and WIN2162-EXIT-TAIL-RIGHT-1124 rows; visible row overlap/joins. Independently confirms reported captured display failure, not byte-delivery cause.
Whole app Terminal3 retains old WIN2162-EXIT-TAIL-RIGHT-1124 and predecessor command intermixed with successor WIN2162-RESTART-SUCCESS-1125/PID14372 after hide/reveal. Direct visual confirmation; cause not established.
Whole app new Terminal3 displays clean WIN2162-UI-REPLACEMENT-1126/PID7936/cwd/42 with no predecessor rows visible.

Protected WSL

Evidence	Screenshot
Requested120×40/same guest after reveal, then89×44
Outer exit7 final marker/output/code7
Codex useful target/reload replies
Child geometry command/exit0 (not outer-PTY proof)
Claude useful target/reload replies
Claude host reply/post-reload
Hidden/reload; chosen-cwd history versus restored root
Useful successor
Usable restored shell; root cwd

Judgment points: one operation owns acquisition and cleanup, durable exit retirement before outcome, proof-bound reply projection.