mirror of
https://github.com/stablyai/orca.git
synced 2026-09-21 08:02:32 +00:00
main
11474
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Jinwoo Hong
|
e6fa905e91 |
fix(i18n): regenerate the runtime-required catalog for the source-control tooltip key (#21942)
#21733 consolidated the source-control tooltips and added auto.components.right.sidebar.SourceControl.527e130b6f to en.json without regenerating en-runtime-required.json, so verify:localization-runtime-catalog fails on main and on every PR's static-analysis job. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
 Neil
|
ac4d6b407a |
fix(devin): skip workspace trust for Orca launches (#21925)
* fix(devin): skip workspace trust in yolo launches * fix(devin): migrate existing workspace trust defaults * fix(devin): persist migrated launch arguments * fix(i18n): include required source control stop label |
||
|
Jinwoo Hong
|
8d42410e01 |
feat(mobile): render the HTML preview on the page in a sealed srcdoc frame (OTA phase C, C7.10 A) (#21862)
* feat(mobile): offer a cancelled top-frame navigation to the shell's opener Both shells cancelled every navigation off their own document in silence: iOS `decidePolicyFor` allowed only `isMainFrame && isDocumentUrl`, Android's `shouldOverrideUrlLoading` dropped anything whose resolved path was not "/". Nothing opened. That is the whole of ruling 29's "if they do not": a user tapping a link inside C7.10's sealed HTML-preview frame reaches the top frame as a navigation request, and the shell was the only thing that could act on it. A cancelled main-frame navigation now reaches JS as `onExternalNavigation` and goes through the same `Linking.openURL` the `externalLink` notify already uses. The scheme list is not restated natively: the native side caps the string and says which frame it came from, and `readBridgeExternalLinkUrl` decides what opens in the half that ships over the air. A subframe navigation is never offered, because that is the sealed preview loading itself. swiftc check: OK (`checkCancelledNavigation` added, the whole suite runs). Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * feat(mobile): render the HTML preview in a sealed srcdoc frame on the page C7.6 gave the page the artifact's source, which is the native component's Source tab and half its job (ruling 8). Ruling 26 makes that debt: the Preview tab comes back as an `<iframe sandbox srcdoc>` inside the page's own document. `srcdoc` rather than a `blob:` URL, and no CSP change at all. Measured on Chromium and WebKit: a `srcdoc` frame has no URL for `frame-src` to match and inherits its embedder's policy instead, so it is admitted under the shipped `frame-src 'none'`, while a `blob:` frame is refused by `frame-src` on both and refused a second time in WebKit by the `frame-ancestors 'none'` it inherits. Two independent fences seal it, and the render check measures each on its own: the sandbox grants neither `allow-scripts` nor `allow-same-origin`, and the inherited `script-src 'self'` refuses the artifact's inline script even when a control arm grants `allow-scripts`. The inherited `img-src` and `font-src 'none'` govern its subresources, against a no-header control where the same three are fetched. `allow-top-navigation-by-user-activation` is the one token granted (ruling 29), so a tapped link becomes one top-frame navigation the shell now opens externally, while a `<meta refresh>`, a form submit, `target="_blank"` and any script-initiated navigation produce none. `lucideBarrelPlugin` is exported from the bundle builder so the check builds the toolbar's icons the way the page does rather than carrying a second shim. config/scripts suite, this file: 14 passed, 0 errors, exit 0. Control runs: a literal `sandbox` in the JSX reds 4, an added `allow-scripts` reds the script fence and the token census. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): pin the preview's sealed frame where the degradation was pinned The three HTML-preview cases in this file described the state ruling 26 retires: no toggle, no frame, the source only. They now pin the frame's shape through the test renderer -- the artifact reaches it as `srcDoc`, the sandbox grants neither `allow-scripts` nor `allow-same-origin`, both toggle positions exist, and Source takes the frame away with it -- and the "never renders the html itself" case becomes "never puts it anywhere but the frame", counted rather than merely absent. What a browser does with that frame stays in the render check, which is the only thing that can answer it. The rich Markdown editor's half is unchanged: it is still the plain field, and item C is a later PR. Two mocks added: `Pressable`/`ScrollView` on the react-native double, because the toggle renders one, and `lucide-react-native`, whose barrel imports a `LucideProvider` its own context module does not export and so does not load under vitest at all. 9 passed, exit 0. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): refuse a link-activated top-frame navigation, even to the document F1, blocking, with F5 and F6 folded in because they are the same decision and splitting them would mean three rewrites of one function. F1: `<a href="/" target="_top">` and `href=""` in an artifact resolve against the embedder's base, so both named the shell's own document URL -- which both shells ALLOWED (iOS `isDocumentUrl`, Android's path `/`). One tap inside the sealed preview reloaded the shell's page: bridge target cleared, load state restarted, page state gone. A navigation a human started is now never allowed, whatever it names; it is offered instead, and `cancelledShellNavigationTarget` drops `orca-mobile-web:` in silence exactly as it drops `/h/other`. The page rewriting its own path carries no gesture and is still allowed. F5: the OFFER is gated on the same gesture, so a top-page meta refresh or a redirect is cancelled and never opened externally. F6: iOS returned early on `shouldPerformDownload` before the offer, so `<a download>` was dead on iOS and opened on Android. The early return goes; a download is refused rather than allowed when nothing started it, and a gesture-started one reaches the opener on both platforms. The allow half and the offer half are now one function per platform (`MobileWebShellNavigationPolicy.verdict`, `mobileWebShellNavigationVerdict`), so they cannot drift. The gesture is the platform's own answer: `.linkActivated` on iOS, `request.hasGesture()` on Android. Native tests, both platforms: document URL + gesture refused and offered; document URL without gesture allowed; foreign + gesture cancelled and offered; foreign without gesture cancelled and silent; download both ways; subframe never offered. swiftc OK; control run with the gesture rule removed exits 133. Gradle MobileWebShellDroppedNavigationTest tests=8 failures=0 errors=0. Also corrected: the screen comment that claimed the document's own reloads reach the handler (they never do), and the prop doc, which now states the gesture rule. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): count own-origin top-frame navigations, and drop the goto cap F2: `page.setDefaultTimeout(4000)` capped `page.goto` at 4 s while every sibling render check uses the 30 s default, so under load the first WebKit cases redded on the navigation rather than on anything they assert. The cap goes; the per-action timeouts that needed to be short are already passed at their call sites. F1's page-side half: the rig now routes the page's own origin as well as the foreign one and counts main-frame navigations to each separately, with two cases pinning that `href="/"` and `href=""` each produce exactly one own-origin top-frame request. Playwright is not the shell, so what these state is the request the shell is handed; refusing it is the native tests' job and the docstring names which ones. The own-origin route is registered after the initial load, because it aborts main-frame navigations and the first `goto` is one. The foreign-tap and meta-refresh cases now also assert zero own-origin navigations, so a fix that merely moved the target would not pass. 16 passed, exit 0, no Errors line. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): wait for the preview frame's own load, never a clock CI read the child frame before its srcdoc committed: frameUrl came back '' and the control arm's script as not yet run. The frame list, the frame's URL and anything read inside it settle at their own moments, and a 900 ms wait reads whichever of them has happened -- on a loaded runner, none. Polls for a child frame at about:srcdoc with its load fired, bounded by the case's own timeout, and an override arm now resolves on the document its srcdoc assignment commits rather than on the assignment. Red-first: with a 2.5 s mount delay standing in for a loaded runner, the paint case failed on both engines before this and all 16 cases pass after. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * docs(mobile): say whose violations the preview rig reads The list is the main frame's: securitypolicyviolation does not cross into a frame, so an empty one says the embedder raised none and says nothing about the artifact's own style, image or font. A listener inside the frame cannot be the fix -- the fence under test is that nothing in the artifact runs. So the comment now claims what the reading supports, and names where the frame's containment is actually measured: the pixel for its inline style, the counting server for its img-src and font-src. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): announce which side of the preview toggle is showing The Preview/Source pair carried a label each and nothing else, so which one was showing lived only in the active background -- invisible to a screen reader on both surfaces. Each button is now a tab carrying its selected state, inside a tablist, and the two files' toolbars stay character-identical so the page and the phone announce the same thing. Red-first: the new case renders both siblings and failed on both for the missing role before this. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): type the WebView mock like the file's other hosts The anti-slop gate refuses a bare `object` parameter. Takes the same shape as the react-native mocks beside it, which pass it. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): allow only the load the shell itself started The document URL was allowed whenever the host reported no gesture, so a navigation the shell never asked for could reload the page out from under the session. Measured against a real WKWebView off-device: a sandboxed subframe navigating the top frame to the document URL arrives as `.other` with no gesture at all, and Chromium's own docs allow hasGesture() to be false for a request a human started. Census first: nothing in the page navigates the top frame -- no location assignment, reload, replace, window.open or form -- the router moves by pushState and replaceState only, so the rule needs no gesture and no page cooperation. Both shells now raise a flag around their own load and drop it at commit, and allow a main-frame navigation only while it is up. Everything else naming the document is refused and never offered, since offering it would send the user out of the app. iOS carries the second discriminator the same probe measured: sourceFrame is the main frame for the shell's own load and the subframe for a subframe's top navigation, so a subframe can never take the allow path. Red-first: the Swift checks and the Kotlin tests were written first and failed to compile against the old signature. 9 Kotlin tests, 54 in the module. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): point the meta-refresh arm at the embedder's own URL The fixture pointed off-origin, so its own-origin assertion could not move whatever the frame did. The new arm refreshes to `/`, which resolves against the embedder's base, and pins zero top-frame requests on a counter the `href="/"` case proves reads 1 in the same rig. It also counts what the frame asks for itself, with a presence control that attributes the fence: with `allow-same-origin` and no policy the same fixture navigates the frame to the embedder's `/`, and with the policy dropped but the product's token kept it navigates nothing, so the opaque origin is what refuses it rather than the CSP. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): read what an action produced, not what a clock allowed The 600 ms after every action is gone. An arm that expects a navigation now returns the moment the route handler records it, with a deadline only so a click that missed its target says so instead of spending the case's timeout. An arm that expects none waits for two painted frames inside the page and one 200 ms drain for the popup queue, which is a browser-process event with no in-page counterpart; the docstring says why that one is bounded. Measured and reported rather than claimed: with the new wait replaced by a no-op every arm still passes, because the reads that follow are each a round trip. It is insurance against the runner load that produced the frame-commit race, not a fix for a failure seen here. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): take the settling branch as a ternary What oxlint's prefer-ternary asks for, and the changed-code gate with it. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): find the preview frame by its element, not its URL CI timed out on all seven preview cases on one engine: the poll waited for a child frame whose URL reads about:srcdoc, and that browser reports an empty URL for a srcdoc frame, so every case ran to its own timeout. The same difference had already shown as `expected '' to be 'about:srcdoc'`. The frame is now the element: waitForSelector('iframe') then contentFrame(), with readiness taken from the fixture's own marker inside it. Nothing compares a frame URL any more -- the paint case reads the element's srcdoc attribute and the absence of src instead, which is what "parsed inside the frame rather than fetched into it" actually means. The one arm whose artifact navigates the frame away says so rather than waiting for a marker that is not coming. Red-first: with the old poll keyed on a URL the browser never reports, both engines time out exactly as CI did; the new wait passes 18/18 with the 2.5 s mount delay still injected. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): make a frame that never becomes ready say what it saw The runner's Chrome read the preview frame's URL as empty where three chromium builds here read about:srcdoc: bundled headless, the headless shell, and --headless=old, all 147. So the difference is not reproducible locally and the next CI run has to carry its own diagnosis. The marker wait is bounded well inside the case timeout, and on expiry it reports the frame's URL, the srcdoc attribute's length and the page's CSP violation list -- which separates a frame the policy refused from one that was merely slow, the two readings that look identical from a timeout. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): run the containment arms the comment only claimed The comment said the fixture navigates nothing with the policy dropped and the product's token kept, but no arm ran it: the control dropped both fences at once. Both single-fence arms exist now, either of which would hold. Measured rather than assumed, and one of them is not what the comment said. The token alone: the navigation never starts, no request, no violation. The policy alone, with allow-same-origin granted: the navigation does start and frame-src refuses it, which the embedder reports as its own violation. The engines differ only in what is left in the frame -- chromium an error page, WebKit the artifact -- so neither is asserted; what is asserted is that the request never reaches the server. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): refuse a download that names the shell's own document The document branch skipped downloads, so `<a href="/" download>` fell through to the offer path carrying the shell's own URL. Harmless in practice, because the opener's scheme list drops it, but it contradicted the policy's own comment and the prop doc, and it left the one URL that must never be offered reaching the boundary. The branch now covers a download too: refused, from either frame, gesture or not, and never offered. A gesture-started download of anything else still reaches the opener. Red-first on both platforms: the Swift checks exited 133 and the Kotlin row failed against the old policy. 10 navigation tests, 55 in the module. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): drop the own-load flag wherever a document ends The flag lived beside the load call and had to remember every ending separately, so iOS missed two: a prop update that fails before it loads, and a renderer that died. Both left it raised, and a navigation to the document URL during that window would have been allowed. It now lives in the load state machine, which every ending already runs through -- a commit, a failure, a dead renderer, a prop update, a reset -- on both platforms, so there is nothing left to remember. The view raises it and reads it, and drops it nowhere. The Android residual is stated in the policy rather than papered over: between loadUrl raising the flag and onPageStarted dropping it, a navigation to the document URL from inside the preview frame would be allowed, because that callback says nothing about which frame asked and no host discriminator exists. It needs a generation switch and a tap in that window; iOS closes the same gap with sourceFrame. Red-first: the new Swift row failed to compile and the Kotlin row with it. 12 load-state tests, 56 in the module. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): spend the own-load flag on the allow, not on the commit The flag stayed raised from the load until didCommit, so a second main-frame action naming the document inside that window was allowed too and replaced the document. WebKit can decide a second action before the first one starts, so the commit is too late to be what spends it. The allow itself spends it now, before the decision goes back, and every ending still drops it for a load that is allowed and never commits. Red-first: the new check composes the machine with the policy -- the seam the flag and the rule meet at -- and failed to compile against the old machine. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): stop raising an own-load flag Android never consults WebViewClient's javadoc, verbatim: "This callback is not called for all page navigations. In particular, this is not called for navigations which the app initiated with loadUrl(): this callback would not serve a purpose in this case, because the app already knows about the navigation." So the flag guarded nothing on this platform and, while raised, was the one thing that could have let a competing request to the document URL through. The view passes isShellLoad = false always now, the machine drops the field it had no raiser for, and the policy comment carries the quote. Nothing reaching that callback is the shell's own load, so nothing naming the document is allowed there at all -- which also closes the generation-switch window the residual named, so that paragraph goes. No red to show: this is a removal, and the behaviour it leaves is the refusal the existing rows already pin. What a device proof must check is stated in the policy instead: a WebView that did route its own load here would have it refused and the load state would sit at loading. 55 tests in the module. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): settle every arm, not only the ones that tap An arm with no action read its counters as soon as the frame's marker appeared, so a zero-delay meta refresh could dispatch after the reading. The arms that pin zero were the ones relying on it. Every arm settles now, and what it settles on is what it expects: the sealed refresh arms take the bounded no-navigation path, and the loose arm waits for a recorded navigation that is neither main-frame nor foreign -- its own frame's -- rather than the main-frame wait it would never satisfy. Red-first: with the settling removed and the refresh moved to 2 s, the loose arm reads 0 on both engines; with it back, 1 on both, the delay still in. A 0.4 s refresh passes either way, which is why the finding was invisible. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): wait for what the artifact's script wrote, not for the element The two-fences control asserts the inline script ran, and the marker element it waited for exists from parse time, so the arm could read window.__ran before the script had touched it. Under a loaded runner that reads 0, which is CI's "expected +0 to be 1" on chromium. Readiness is now per-arm: 'script' waits for the script's own write, 'load' for the arm whose artifact navigates the frame away, 'artifact' for the rest. Red-first: with the inline script's write delayed 1.5 s, the old arm fails on both engines with that exact message and the new one passes, delay still in. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): bound the rig's waits by the case timeout and nothing else Two inner deadlines, 20 s and 15 s, were racing the outer one they sit inside, so a slow runner could fail a case on a number this file picked rather than on the one the case declares. Both now run to vitest's own `ctx.signal`, which aborts when the case times out. On abort the rig prints its reading -- the frame's URL, the srcdoc length, the violation list, or the navigations it did record -- and lets the case fail as the timeout it is. Nothing is rethrown from that path: a rejection raised after vitest has given up on a case has nobody left to catch it, and an unhandled one fails a run whose every test passed. Red-first: with the marker selector pointed at an element that never appears and the case timeout cut to 8 s, the diagnostic prints and the case fails as `Test timed out in 8000ms` rather than hanging in silence. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): ask a stuck preview frame everything it can still answer The old diagnostic said only that a frame never parsed, and its violation list was the top document's -- securitypolicyviolation does not cross frames, so it said nothing about what the frame itself refused. It now prints the browser version, the arm it came from, the iframe element's srcdoc length and sandbox, contentDocument.readyState and contentWindow.href (which answer for a same-origin arm and report `refused` for an opaque one, so the arm's own origin is in the log), and every Playwright frame with its url, name, readyState, body length, marker presence, window.__ran and its own violations. Per frame, because the page's init script installs the collector in every frame -- measured on both engines -- and CDP evaluates inside an opaque frame whose scripts are blocked. Two corrections that the local probes forced. The reading is sampled while waiting and printed from the last sample: read at the abort it lost its race with vitest's teardown and printed nothing at all. And two arms had never been given the case's signal, so their waits could not be bounded or diagnosed. The diagnosis moves to its own module because the test file is at its line limit, and because the bound and the reading it prints are one thing. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): build a widened control frame instead of relaxing a live one A live frame cannot be relaxed. Sandbox flags are fixed on a browsing context when it is created, and Chrome 152 keeps the original ones through a srcdoc reassignment while still parsing the new document -- so the control arms that widened the product's own frame stayed sealed on the runner, and CI read a script that never ran and a refresh that never navigated. Chromium 147 here honours the relaxation, which is why it passed locally for a year of runs. The override now clones the element, sets the sandbox on the clone, gives it the artifact and replaces the product's frame with it, so the widened flags are there from creation -- the way the product does it, since React sets the attribute before insertion and never after. The product's own arms are untouched: a null override still returns immediately. And the control can no longer pass for the wrong reason on any engine. The header-keeping arm now reads the violation raised inside the frame: a script-src refusal can only happen if the sandbox let the script start, so it separates "the policy held" from "the frame was never widened", which the old arm could not. The loose arm pins an empty list beside it, the sealed arm pins an empty one too, and those three readings are the whole fence story. The violations come from each frame's own collector, because the embedder never sees them. Two diagnostic repairs the local probes forced: the browser version is read once at open, since asking at the abort printed "browser unknown" in the CI log this exists for, and the reading is sampled immediately as well as every five seconds, since a wait that only prints "no reading was taken" says nothing. Red-first: with the widening disabled, both engines fail exactly as CI did -- 180 s timeouts on the script arm -- and the diagnostic names the arm, the version and the sandbox it actually had. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): put the toggle's selected state where a browser reads it CodeRabbit is right, and the browser says so: react-native-web's createDOMProps never reads accessibilityState, so on the page the tab pair emitted role="tab" and no aria-selected at all. The test renderer could not see it, because it reports the props the component was handed rather than the DOM they become. Both siblings carry aria-selected beside accessibilityState now -- the phone's screen reader takes the latter, the browser the former -- and the toolbars stay character-identical. Red-first, in a real browser on both engines: the rig now reads every [role="tab"] element's aria-selected before and after the tap, and it read null for both positions before this line existed. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
 Jinjing
|
1b77838d1c |
Consolidate source control tooltips to eliminate redundant hover text (#21733)
* refactor(source-control): consolidate tooltips to avoid duplicates - Remove native title attributes from buttons that also render Radix tooltips - Introduce PrimaryActionTooltip wrapper that decides whether to show a tooltip based on context - Hide pure repeats: enabled Stage All and Create PR have no tooltip since the label already states the action - Show tooltips only when they add information: disabled reasons, commit shortcut, and remote counts * Show Create PR intent tooltip to explain the prepare step The Create PR intent label doesn't convey that clicking it stages, commits, and pushes before opening the PR dialog. Keep the tooltip to surface this multi-step operation that users might not expect. |
||
|
Jinwoo Hong
|
5e5d731870 |
test(config): assert the target route's chunk was never fetched, not that no chunk arrived after the tap (#21922)
* test(config): name the chunk the handoff check says is absent The three handed-off cases snapshotted the loaded scripts the moment the page returned and then asserted nothing new arrived after the click. The opener's own route chunk can still be in flight at that moment, so under full-suite load it lands afterwards, counts as new, and reds a case whose rule held: the notify was posted and the document had not moved. They assert what the comment always claimed instead -- the target route's own chunk was never fetched -- resolved from the build's route-to-chunk map, since the bundler hashes chunk names and there is nothing in a URL to recognise a route by. Being about one named chunk over the whole run rather than a delta, it cannot be raced by an unrelated arrival. An absence needs a presence precondition, so `beforeAll` fails if that route has no chunk; without it a typo would pass all three. Red-first control: with the coverage test in `route-handoff.web.ts` forced to keep the hop local, the chunk assertion alone fails and names the tasks chunk it fetched. Found by the C7.5b lane. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(config): judge the absent chunk on every status, not only 200 CodeRabbit: the three absence assertions filtered the 200-only list, so a request for the tasks chunk answered 404 or 500 left no trace in it and the assertion passed on a fetch that did happen -- the one shape where a page that tried to render the target reads as a page that never asked. They read `jsResponses` now, which carries every JavaScript response with its status, and a request for that chunk fails the case whatever came back. The 200-only list had no other reader, so it is gone rather than left as a second collection to pick the wrong one from. Re-proved on this form, with the coverage test forced to keep the hop local: the chunk assertion alone reds and now prints the status beside the path. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
|
Neil
|
37dbb6cf6e |
fix(windows): prune the unpatched conpty prebuild by header, not host arch (#20048)
* fix(windows): prune the unpatched conpty prebuild by header, not host arch `prunePackagedNodePty` deleted the published `prebuilds/win32-<arch>/conpty.node` only when `electronArch === process.arch`. That proxy stood in for "build/Release holds an addon this slice can load", and it is false for the arm64 slice cross-built on an x64 Windows host — a rebuild that DOES emit a correct arm64 addon. That slice shipped the unpatched prebuild. Nothing loads it today: `verifyPackagedConptyBreakawayMarker` resolves the addon the way node-pty's loader does, so the patched `build/Release` wins and the release passes correctly. But the loader swallows every require failure and falls through, so an AV quarantine or a missing dependency on `build/Release` hands the pane to that unpatched prebuild — the silent downgrade the gate exists to close, with the binary still sitting in the package. Read the PE `Machine` field instead of guessing, reusing `readPeMachine` from the verifier's `windows-pe-machine.cjs` so prune and verifier ask one question. A missing, truncated or non-PE `build/Release` reads as unloadable and keeps the prebuild, which is what the true cross-host case needs: packaging Windows from macOS leaves no Windows binary in `build/Release`, and removing the prebuild there would leave the package with no ConPTY at all. Mutation-proven: restoring the `electronArch === process.arch` guard fails exactly the two new rows in packaged-node-pty-prebuild-prune. * docs(windows): note the cross-arch conpty slice is real but not yet built |
||
|
Neil
|
4e45fd04a1 |
docs: drop the removed WeChat group 8 QR from the translated READMEs (#21936)
#21927 removed docs/assets/wechat-qr-group8.jpg and updated README.md, but the fr, ko and zh-CN translations still referenced it. The README local-link check fails on main today, so every PR run goes red on the root directory guard until this lands. Mirrors what #21927 did to README.md: the group 8 image is dropped and the copy now points at group 9 only. |
||
 Brennan Benson
|
afb618f2b3 |
refactor(agent-status): drop two superseded Codex attention workarounds (#21844)
* refactor(agent-status): drop two superseded Codex attention workarounds Codex fires its PermissionRequest hook as decider #1, before its own auto-reviewer and before the user, so the event never meant "a human is blocked". #21389 fixed that at the source: the execution host reads the turn's approvals_reviewer from the rollout at write time and keeps a reviewer-owned approval in `working`. Two older reader-side workarounds for the same bug are now redundant. The launch-argument suppressor guessed auto-approve mode by string-matching the launch args, then dropped the status row in the reader. It only matched Codex's bypass flag, and under that flag Codex's approval policy is `Never`, which takes the Skip path and fires no PermissionRequest at all. When the user turns on "Approve for me" inside a live session the args never change, so it never fired for the actually-reported case either. The Codex-only 1.5s notification quiet window could not do its job: measured auto-reviews take 3-20s and a human can answer in under a second, so no fixed constant separates them. Its deferred callback also re-checked liveness and returned without notifying, so a genuine prompt whose pane went non-live inside the window was dropped rather than delayed. Codex now notifies synchronously like every other agent. Also types the coordinator's completion state from the controller's exported CompletionState instead of asserting each field, which the changed-lines casting gate required once those lines moved. * fix(agent-status): settle transient process-exit evidence |
||
|
Jinwoo Hong
|
4a3a32206d |
refactor(mobile): the terminal document is a function of its host (OTA phase C, C7.5b) (#21859)
* test(mobile): pin the terminal WebView document byte for byte The document is already pinned as a digest, which says whether the emitted bytes moved and nothing about where. C7.1 moves the hand-written script inside it into modules the web page can import and rebuilds the document from them, and the claim that has to hold through every one of those commits is that the native screen kept the document it had. A digest cannot be the instrument for that: it fails as two hexadecimal strings. So the document is also committed as itself. The fixture is generated by `scripts/build-terminal-document-fixture.mjs`, never pasted, and the test rebuilds the comparison through that script's own substitution rather than restating it, so a fixture written by one rule and read by another cannot agree with itself. The generated xterm engine is stored as two placeholders. It is already covered by the digest test, postinstall regenerates it from whatever xterm the lockfile holds, and inlining it would put 612 KiB of vendored bytes into the file whose job is to isolate hand-written changes. Two further cases keep that from becoming a hole: the placeholders must each appear exactly once and the engine must not appear at all, and the restored document must equal the real one. Regenerating the fixture is a review event. It is only correct when the emitted document was meant to change, and the diff in that commit is the evidence. Red-first: flipping one character inside a comment in `write-queue.ts` fails both identity cases with a one-line diff naming the comment, where the digest test reports a hash. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): compare two terminal documents as programs, not as bytes The C7.1 flip commit moves the document's 57 reassigned variables onto a scope object, because a variable assigned across ES modules is a syntax error, and every read and write of them gains a qualifier. The ruling asks that the review of that commit be a test rather than a 515-line read. This is that test's instrument. It cannot be a byte comparison. Once the script's source is modules, `oxfmt` owns its style, and the repository's style has no semicolons where the hand-written document has one on nearly every line. A byte diff would therefore be dominated by changes that are not the refactor, which is the opposite of what the reviewer needs. So the comparison is over tokens: semicolons are excluded for the same reason they moved, comments never reach the stream, and one difference is allowed — `name` becoming `<qualifier>.name`, three tokens for one — which it counts and reports. It is stricter than "it still runs": a reordered statement, a changed literal, a dropped operator, a renamed local and a qualifier under the wrong object name all diverge, each reported with the token index and both sides. Acorn carries `value` on its tokens but does not declare it, so the field is read through a narrowing check rather than asserted onto the declared type. Red-first, by mutation: dropping the qualifier-name check fails the case that names it; removing the leftover-token check fails the dropped- and added-statement cases; treating semicolons as significant fails the three cases that depend on ignoring them. The acceptance case runs on the real 2,758-line script rather than on a fixture, so the instrument is known to survive everything the document actually contains. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): count each normalisation the move makes, separately Measured while extracting the first group: the document's ES5 style is not a style this repository's own rules permit. `curly` braces 279 brace-less if/else/for/while bodies, `no-unused-vars` unbinds 38 catch clauses, and 446 `var` declarators become `const`, `let` or a scope field. Those rewrites land before the qualifier is considered at all, so "the qualifier and nothing else" was never reachable once the source is a linted module. The comparison now allows exactly four classes and counts each on its own: a reference that gained the qualifier, a declaration that moved onto the scope object, a `var` that only changed keyword, a body that gained braces, and a catch clause that lost its binding. Separate counters rather than a total, because the flip commit pins each number and a total would let one class absorb another — which is the drift the pin exists to catch. The two `var` classes partition the 446, and the qualifier's 641 sites partition into references that kept their declaration and declarations that moved. Two ordering facts the cases pin. The catch rule is tried before the brace rule, or the inserted-brace rule eats the `{` that follows `catch` and the streams never resynchronise. A body braced at the very end leaves its closing brace after the baseline has run out, so trailing closes are absorbed after the walk rather than reported as a length difference. Everything outside the four classes still refuses with the token index and both sides: a changed literal, a dropped operator, a reordered pair, a renamed local, a qualifier under another object's name, a brace opened and never closed, and a brace closed where none was opened. Red-first, by mutation: disabling the catch rule, disabling the trailing-brace absorption, folding scope-field declarations into plain references, and not counting brace insertions each fail exactly the case that covers them. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * refactor(mobile): make the mouse-report cell a module the page can import The first of the twelve groups the document already names. `*-injected.ts` has been splicing JS strings into the document for a while, and tests evaluate those strings, so the one-source-two-consumers shape is already there; what is missing is that a string cannot be imported by the web page, typechecked, or linted. This turns one of them into a module and adds the generator that puts it back into the document. The generator is a transform, not a bundle: a bundler orders its output by the dependency graph, and the document's order is part of what the equivalence test holds fixed. Imports are dropped rather than resolved, because inside the document every name is already in scope — that is what the single IIFE means — and `document-externals.ts` declares the names whose groups have not moved yet and emits nothing at all. esbuild prints an ESM module's exports as a trailing block, so that block is dropped whole rather than by its keyword; leaving the keyword behind would put a bare block statement in the document. Both sides of the comparison now go through that same printer before being read. Otherwise every choice the printer makes — semicolons, property shorthand, quote style — reads as a difference in the program when it is a difference in who typed it, and each would need its own rule. A script that does not parse is reported as a refusal naming its side, not thrown. `let` is contextual outside strict mode, so acorn reports it as a name and not as a keyword; without that the var-to-let rewrite the linter performs would be refused on every reassigned local. The group's counts are pinned exactly: nine references gained the qualifier (`term` seven times, `panX` and `panY` once each), nine locals became `const` or `let`, thirteen one-statement `if` bodies gained braces, no declaration moved onto the scope object and no catch clause lost a binding. The document is untouched, so the byte pin from |
||
|
Neil
|
b2fe56def9 |
fix(worktree): recognise the Windows profile through WSL's drvfs view (#20051)
* fix(worktree): recognise the Windows profile through WSL's drvfs view `/mnt/<letter>` under a WSL UNC alias is the distro's drvfs mount of a Windows volume, so `\\wsl.localhost\Ubuntu\mnt\c\Users\bob` is `C:\Users\bob` wearing a Linux spelling. The Windows-profile rule excludes every WSL UNC path by design (the aliases normally front a Linux filesystem) and the POSIX shapes never match a `/mnt/...` tail, so that path fell through both and read back as deletable. The spelling is producible by the product: `resolveWslRepoWorktreeBasePath` maps a `/mnt/c/...` worktree base against a WSL repo into exactly this UNC form, and `getWslFilesystemBoundaryDistro` already treats it as the drvfs crossing. A drvfs tail now takes the Windows rule on its drive form, via the existing `toWindowsWslDrivePath`. Scoped to the UNC branch, where `parseWslUncPath` has proven the path is a WSL alias — a plain Linux host's `/mnt/c/...` is untouched. The lowercase-only `/mnt` match is deliberate: `/MNT` is an ordinary case-sensitive Linux directory, never the automount. * fix(worktree): refuse the drvfs volume root and the automount under a WSL UNC alias `\\wsl.localhost\Ubuntu\mnt\c` is the whole C: volume and `\\wsl.localhost\Ubuntu\mnt` holds every drvfs volume. Neither is caught by the root check in `isDangerousWorktreeRemovalPath` (their win32 root is the distro share) nor by the Windows-profile rule on the drive form (`C:\` is not `C:\Users`), so both read as deletable. Measured on a Windows 11 host with WSL2: `rm -rf` inside the distro on the `/mnt/c` spelling deletes on the Windows drive. |
||
|
Brennan Benson
|
4da3a95d50 |
fix(native-chat): scope a tool row's hover reveal to that row (#21918)
Hovering one tool call in a chat turn revealed the expand chevron on every other row in the same message at once, so the whole message lit up and nothing said which row the click would open. The rows were reading a hover they do not own. Tailwind's unnamed `group-hover:` is not nearest-ancestor scoped — it compiles to `:is(:where(.group):hover *)`, which matches a hover on ANY `.group` ancestor. `NativeChatMessageRow` wraps the whole assistant message in a bare `group` for its own copy/timestamp reveal, so every collapsible row nested inside it answered to that wrapper as well as to itself. Each row now names its own group — `group/tool-line`, `group/tool-run`, `group/subagent-run`, `group/diff-card` — which compiles to `:is(:where(.group\/tool-line):hover *)` and reaches that row alone. The message-row reveal is left bare on purpose: its copy button and timestamp are meant to answer to a hover anywhere in the message. `NativeChatDiffCard` is included for the same defect, not as extra scope: its verb label was brightening on any hover in the message. |
||
|
Neil
|
8cf1c8594e | docs(opencode2): clarify quick command delivery (#21929) | ||
|
Jinjing
|
3bb10e3f0c | docs: remove obsolete WeChat group 8 QR code (#21927) | ||
|
Jinjing
|
e497ef36f2 | docs: update WeChat group 9 QR code (#21926) | ||
|
Neil
|
70c4f20466 |
fix(opencode2): auto-submit quick command prompts
OpenCode2 quick commands now submit through the ready-state delivery path. Focused regression coverage and full CI pass. |
||
|
Neil
|
98299d879b |
fix(terminal): persist a parked remote pane's scrollback across a hard restart (#21295) (#21367)
* fix(terminal): route a parked pane's scrollback patch to the remote host's partition A park capture changes only terminalLayoutsByTabId, so its debounced session patch carries no tabsByWorktree. splitWorkspaceSessionByHost built its tab->worktree index from the patch alone, resolved nothing, and routed every layout to the 'local' partition, where main's pruneLocalTerminalScrollbackBuffers strips scrollback it cannot attribute to a remote worktree. The remote host's runtime:<id> partition never received the capture, so anything parked since the last clean checkpoint was lost on a crash, SIGKILL, or a forced kill during an app update (#21295). Route tab-keyed patch fields with the renderer's live tab catalogs as a fallback when the payload names no tab rows. Payload rows still win, so full-payload writes are byte-identical. Once routed to runtime:<id>, main merges the partition's own prior tabsByWorktree and the prune preserves. Proven by tests/e2e/paired-remote-terminal-parked-scrollback-restart.spec.ts: a hard kill (no checkpoint) then relaunch, asserting the capture is in the remote host's partition on disk. Mutation: reverting the routing fix turns that assertion red and fails the 3 catalog-dependent unit routing tests. (cherry picked from commit |
||
|
Neil
|
f492054bf0 |
fix(runtime): an outage is not a handle-gap verdict (#20059)
* fix(runtime): an outage is not a handle-gap verdict
The per-pane handle-gap wait releases at a 15s deadline and records that
expiry as a verdict, which authorises the sleeping-agent resume. The
connection generation was the only thing voiding that verdict, and a plain
disconnect never advances it — runtime-status.ts advances on the reconnect,
under a new runtime id. So a network drop mid-turn expired the wait with a
generation that still matched, and the replay forked a second `--resume`
onto the transcript the host was still writing: #19735 through the
disconnect door.
Suppress the verdict while the client positively knows it is out of contact,
reusing the shared runtime-host connection derivation. The waiter still
releases and re-parks, so contact returning gets a full fresh budget and the
pane is still decided on real silence.
Not redundant with the landed-handle drain that follows this commit, nor with
the read-time pane identity from adv2-skew (
|
||
|
Neil
|
4feca6baa1 |
Keep new worktree dialog actions visible while scrolling (#21915)
* Keep new worktree dialog actions outside scrolling content * Trigger missing PR checks |
||
  SahilZ0810andNeil
|
9324bb8137 |
fix(editor): preserve Markdown preview when following wiki links (#19790)
* fix(editor): preserve preview when following wiki document links * test(editor): avoid cast in markdown navigation fixture --------- Co-authored-by: Neil <neil@stably.ai> |
||
|
Neil
|
2872c3fccc |
fix(claude): prefill continuation context for manual submission (#21912)
* fix(claude): prefill continuation context for manual submission * fix(agents): force draft paste when inline prefill falls back |
||
|
Jinwoo Hong
|
5d13a70ea3 |
fix(mobile): keep an in-page hop local only when the session's grants cover it (OTA phase C, C2.9) (#21723)
* feat(mobile): carry what each page route declared in init (OTA phase C, C2.9) The page decides an in-page hop from `init.pageRoutes`, which says which patterns this shell would render and nothing about what each one costs. So a push kept local on the strength of the pattern alone runs the target under the opener's grants — which is how the tasks page is reached from the wide-layout sidebar without `native.clipboard.write`, and why its copy actions refuse silently. `init` now also carries `pageRouteGrants`, the manifest's own route/grant pairs, from the manifest the shell already holds. Optional in both directions: an older shell omits it and an older page ignores it, and a page that receives none keeps today's rule. No new frame kind, no cap change, no protocol bump. The grammar is the manifest's, imported rather than restated (`MobileWebBundleGrantNameSchema`, now exported for this), so a grant name the bundle could not have declared cannot reach the page through this field either. The host validates the pairs before it builds the frame and refuses the session when they fail, for the reason it already refuses a malformed route: an `init` the page would reject whole is worse than no session at all. Two files were at their line ceiling and are split rather than bumped. The pairs schema moves to `bridge-page-route-grants.ts`, which is read by both the envelope and the host, so it belonged in one place anyway. In the session reducer the three sites that each spelled out "patterns, their grants, this route's grants" become one `routeViewOf`; that is a net reduction and removes the fourth spelling before it is written. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): keep a hop local only when the session's grants cover it The rule the page was using is "the shell would render this pattern", and that is not the question. Grants are resolved once, from the route the shell opened, so a push kept local runs the target under the opener's list. On a wide layout the sidebar renders beside every `/h` route and pushes `/h/<id>/tasks` through this seam, so from the worktree list, agent history or the files pages the tasks page ran without `native.clipboard.write` and its copy actions refused with nothing on screen to say why. `servedHere` now means served here *and* covered: the target's declared grants must be a subset of this session's. An uncovered page route is handed to the shell exactly like a non-page route, and the shell opens it as its own session with its own grants — which is the mechanism that already exists, rather than a new one. Three answers, not two, because an absent field is not an empty one. A shell that sent no pairs keeps the old behaviour: `null` is "nobody told me", and an older shell has to keep working. A target the shell lists but names no entry for is *not* covered — the page cannot justify that hop, so it hands it over rather than guessing in the direction that loses grants. This is C3.1's explorer ⊇ preview finding without its pairwise pin: that hop is covered by this rule and stays local, and the rule scales to the sidebar, which reaches every route and which no pairwise list can keep up with. Red first on the two cases only the new rule answers; the other four are the regression guards and passed before and after. Two whole-session assertions gained `pageRouteGrants: null`, which is what the reader now returns. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): prove the sidebar hop in a browser, under the session's own grants The unit tests pin the decision; only a browser shows the control exists, is reachable at the viewport where the sidebar renders, and that the document does not move when the hop is handed over. Four cases on the shared harness, which now forwards `pageRouteGrants` (omitted when a caller names none, because an absent field is not an empty one and the page reads the difference). - Wide, session without `native.clipboard.write`: tapping Tasks posts exactly one `navigate` notify, the document stays on the worktree list, and **no new chunk is fetched** — which is what says the page did not quietly render tasks under the wrong grants. - Wide, same tap with the grant added: no notify, the document moves to `/tasks`. Without this the first case would pass on a page that simply never navigates. - Wide, shell sending no pairs at all: the old behaviour, local. An older shell must not start handing every hop over on a field nobody sent. - Narrow: asserts the absence rather than a tap. `app/h/_layout.tsx` renders the sidebar only on a wide layout, and only that header branch labels its Accounts and Tasks controls; the narrow header's are unlabelled pressables. So the hop does not exist at that viewport, and `getByLabel('Tasks')` finding nothing is the honest assertion. That unlabelled narrow header is a real accessibility gap and is not this lane's to fix. Registered in `pr.yml`'s `mobile_web_app` job beside the other render checks. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): census the in-page hops a session's grants cannot cover The rule landed in the commit before this one decides each hop; this says which hops those are, so a route's grants growing — or a new push between two page routes — shows up here rather than as a verb that silently refuses on a device. Openers are every page route, not the one that happens to push. On a wide layout `app/h/_layout.tsx` renders the worktree-list sidebar beside every `/h` route and its header pushes tasks, which is exactly why a pairwise pin is the wrong shape: the sidebar reaches everything, so the census has to be the cross product of what the manifest declares against what the source actually builds. Targets come from the hrefs the app builds, read out of `mobile/src` and `mobile/app` and reduced to route patterns, so a hop nobody writes is not pinned and a hop someone adds is. A presence case asserts the sidebar's tasks push is among them, because a census that stopped finding hops would go quietly green. Two hops are pinned as handed off today, both into tasks, which is the only route declaring more than `navigate` and `storage`. A third case asserts the other half of the rule on the manifest: a target asking for no more than its opener stays in the document. Checked that it discriminates rather than assuming: widening the worktree list's grants to cover tasks fails the pin, and restoring them passes it. **No pin was deleted.** The brief expected C3.1's pairwise explorer/preview pin to be replaced here, but C3.1 is not on this base — `MOBILE_WEB_PAGE_ROUTES` has three routes and no `files` entry, so there is nothing to remove. When C3.1 lands, its pin is this census's to subsume. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): drop an unused import from the hop census `statSync` was imported and never used; `oxlint` fails it. My error: I committed the census on a green test run without waiting for lint, the same order mistake I made earlier in this lane. Fixed forward rather than amended, because the lane forbids rewriting a commit that exists. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): fold C3.1's pairwise grant pin into the hop census C3.1 landed while this branch was open, and it brought the case this lane generalises: the explorer pushes to its own preview, that push stays in the document, so the preview runs under the explorer's grants. Its pin asserted that one pair by name. The census now covers it as a consequence rather than a rule. With the files routes in the manifest the cross product finds six more hops the session cannot cover — the sidebar into files from the worktree list and from agent history, and both files routes into tasks — and it does **not** find explorer → preview, because the preview declares no more than the explorer. That absence is the pairwise pin, derived. So the pairwise block is deleted, with its import. The rest of that file stays: its external-link seam checks and its clipboard-absence control are about what the files closure contains, which this census says nothing about. Checked the extended census still discriminates: granting the explorer `native.clipboard.write` fails the pin, restoring it passes. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): prove the sidebar hop from a files route, not only the worktree list The defect is not "the worktree list pushes tasks". On a wide layout the sidebar renders beside every `/h` route, so the same hop exists from the files explorer, whose session carries `externalLink` but not `native.clipboard.write`. One opener proving the rule would have left the general case to inference, which is the inference C3.1's pairwise pin already made once. Opened on `/h/<id>/files/<wt>` with the files route's own grants, the sidebar's Tasks control posts exactly one `navigate` notify, the document stays on the files route, and no new chunk is fetched. The harness helper now takes the route and the text to wait for, so a case can open on something other than the worktree list without a second copy of it. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): make the render helper wait on the text its caller named The `awaitText` parameter I added in the commit before this one was never wired into the wait, so it was dead and `oxlint` failed it. The case still passed, because the files route renders the host name in its sidebar and that is what the helper was still waiting on — which is exactly the kind of accident a dead parameter hides. Third time in this lane I have committed on a green test run before lint finished. Fixed forward, not amended. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): carry route grants through the download path `onManifestRead`'s download branch set `pageRoutes` and `routeGrants` from the new manifest and dropped `pageRouteGrants`; nothing downstream recomputes it, so every first install and every OTA update reached `ready` with the default or the previous generation's pairs. The page then read each target as listed-with-no- entry and handed off every in-page hop. `routeViewOf` moves to `page-route-policy.ts`, beside the two functions it calls, to keep the reducer under its line cap without a bump; its stale neighbouring comment, which described a filter that moved into it, goes. Red first: the cold-cache and generation-change cases failed, the cached-hit case already passed. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): derive census targets from navigation call sites The reachability filter was inert. Harvesting every `/h/${…}` template caught the five screens that declare their own mount pathname, two `pathname ===` comparisons and the route template types, so every declared route was reachable through its own mount: the pinned table was the all-pairs one, eight hops with the filter and eight without. Targets now come from the arguments of `router`/`navigation` `push`, `replace` and `navigate`, and of `navigateFromHostList`; mounts, comparisons and types are excluded by construction because they are not navigation arguments. Two real hops are not written as a literal, so a local binding or a call is followed one step to the function that returns the pathname: the files explorer is pushed as `{ pathname: descriptor.pathname }` and the preview as `push(createMobileFilePreviewHref(...))`. A call site whose target cannot be read is returned rather than dropped. Derived patterns go from 11 to 10; the pinned table stays at eight because all five page routes are genuinely pushed to. What changes is that the filter now discriminates: deleting the header's two tasks pushes reds the presence case and drops the four `-> tasks` rows from the pin, where the old derivation stayed green on the same deletion because `app/h/[hostId]/tasks.tsx` still declared the pathname. A push added at a real call site appears in the set. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): restore the preview-declares-something guard The pairwise pin this case replaced asserted the preview declares at least one grant before asserting the explorer covers them all; without it two empty lists satisfy the subset check and a route that lost its grants passes. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * docs(mobile): describe the route list under the handoff rule Two passages described the world before this PR: the explorer's note said the census pins its pair with the preview, and a closing paragraph left the sidebar's tasks hop open for a later PR. This is that PR. Covering the preview now buys the in-document hop rather than making it correct, an uncovered target is handed to the shell and reopened under its own grants, and the census reads the explorer to preview relation off this list rather than pinning it by name. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): mirror the manifest's tasks grants in both fixtures CodeRabbit on #21723: both fixtures declared the tasks route as `navigate`, `storage`, `native.clipboard.write` while the manifest also declares `externalLink`, so no covered-session case ever required it. Both now mirror the manifest's four, and the covered sessions hold them. That alone does not make an `externalLink`-blind rule fail, since those sessions hold every grant either way, so the unit suite gains the case that does: a session holding the clipboard but not `externalLink` must still hand the hop off. Mutating the rule to treat `externalLink` as always held reds that one case and no other. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): make a stalled hop name its own cause Both waits for the hop to land read as a bare 30 s timeout when it does not. The CI failure that sent this file back was a `TypeError` inside React Navigation that blanked the document, and it was invisible here because the error assertions run after a wait that never returns. The wait now throws with the page's own account: the pathname it stayed on, the collected page and console errors, the `navigate` notifies posted, the first 300 characters of the body, and every `.js` response since the click with its status. The response listener records every script answer rather than only the 200s, so a chunk the navigation waits on can be seen failing; the 200-only list the no-new-chunk assertions read is unchanged, as is everything the five cases assert. Kept in this file because no other render file waits on the pathname moving. Proved by mutating the rule to hand every hop off: the covered case fails naming the pathname it stayed on, an empty error list, the notify it posted and no scripts since the click — which is the handoff signature, distinct from the crash signature CI saw. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): aim the narrow hop at the control C2.10 named The narrow case asserted the absence of a labelled Tasks control, which was true only because the narrow toolbar carried no accessibility props. C2.10 gave it the wide sibling's role and label, so the assertion was red on the merge and, worse, the rule this file is about went unproven on the branch the phone actually presses. It taps that control now: at 390 px there is exactly one, and the tap posts exactly one navigate notify for the tasks route while the document stays on the worktree list and fetches no new chunk. Red first against the merged header (count 1, expected 0); with the session given native.clipboard.write the hop goes local and the case reds, which is what says the assertions discriminate. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * docs(mobile): drop the handoff predicate's contradicted one-liner The pre-C2.9 summary said the answer is whether this document renders the target, which is exactly the claim the block comment below it replaced: the predicate now also requires the target's grants to be covered. Two doc comments on one declaration, the first of them wrong. Comment only; the 35 handoff cases are unchanged and green. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): assert which field a route refusal blames The host builds `pageRouteGrants: <issue>` so a refusal says which of the two checked inputs failed, and nothing read it: the case counted refusals, so a host that reported the route's own verdict for a malformed pair would have stayed green while sending whoever reads the refusal to a pathname that was never the problem. The case pins the prefix, a non-empty issue behind it, and that the diagnostic and the callback carry the same string. The control is an opener that fails the other way: a malformed route reports its own issue and does not take this prefix, without which the pin would hold on any reason at all. Red first with the field branch dropped from the reason: the prefix assertion fails and the control stays green. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * refactor(mobile): stop exporting the route filter the reducer stopped calling `implementedPageRouteEntries` and `implementedPageRoutes` were the reducer's two ways in before it moved to `routeViewOf`. The entries form had no caller anywhere afterwards and the patterns form had only this test, so the module's public surface advertised two functions no product code reaches. Both are module-local now; the surface is `matchesRoutePattern`, `pageRendersRoute`, `grantsForRoute`, `routeViewOf` and the grant list. The test reads the same list through `routeViewOf(...).pageRoutes`, which is the reducer's own view of it, so no assertion changed and no export is kept for a test. Red first: with both un-exported and the test untouched, seven cases fail with `implementedPageRoutes is not a function`; routed through the view all nineteen pass. Still discriminating, as a control: with the grant filter dropped from the entries helper, four of them fail. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): route the merged haptics cases through the policy view PR E's two haptics cases arrived with the merge calling `implementedPageRoutes`, which this branch had already made module-local, so the merged file was red with `implementedPageRoutes is not defined` on both of them. They read the same list through `pageRoutesOf`, the view the rest of the file already uses, so neither assertion changes. PR E's paragraph named that function for the filter it describes; the filter now sits in the entries helper the view is built on, so the sentence says that instead of naming a function the reader cannot see. Red: the two cases above on the merge. Green: all 21, PR E's two included. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): mirror the haptics token in every handoff fixture PR E put `haptics` on all five manifest routes, and these fixtures still carried the pre-E grant lists: tasks with four grants where the manifest now declares five. A fixture that is short the same token on both sides of the subset check agrees with the rule by accident, and would have gone on agreeing after the token stopped being universal. The pairs mirror the manifest now, and each session carries what its opener route would actually be granted, since the host narrows a route's declared grants to what the shell implements and the shell implements the token. Red first, with the token added to the pairs alone: the two covered-hop cases flip to handed-off, `stays in this document when the session already covers the target` and `keeps the hop in the document when the session covers tasks`. Green once the sessions carry it, 35 and 5. The hop census needed nothing: it reads `MOBILE_WEB_PAGE_ROUTES` itself. Measured there, all 5 routes declare the token and it is the missing grant in 0 of the 8 uncovered pairs, so it cannot decide a hop and the rule still reads only `pageRouteGrants`. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(config): count C2.9's two bridge modules in the session route closure #21908 recorded this pin at 4,324 for the haptics notify module. C2.9 adds two more that the same closure reaches: the page-route-grants schema and the manifest contract whose grant grammar it imports rather than restates, both pulled in by `bridge-envelope.ts`, which the page reads to parse `init`. Named in the docstring beside #21908's sentence rather than folded into its number, because the three modules arrived from two PRs and a single count with one reason invites the next author to assume the rest. Red first against 4,324: expected 4,326. Measured on this head, not inferred -- a control worktree at pristine main gives 4,324, so the two are this branch's. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
|
Neil
|
c74ba15f31 |
fix(claude): stream the provider history window (#21742)
* fix(claude): stream the provider history window An oversized Claude transcript made restart reconciliation unresolvable: readClaudeProviderHistoryWindow buffered the whole project JSONL, so a file past the 16 MiB bound returned an inconsistent boundary — the one answer the reconciler can never act on — and a file just under it still went resident. The window now reuses the streaming primitives from #21024 instead of a whole-file read. Two bounded passes run over ONE pinned descriptor and size: the graph pass builds the branch proof (uuid/parentUuid only), and the replay pass hands back the first record per chain uuid, fingerprinted on the spot. A repair appended mid-read re-runs BOTH passes at the grown size, so the replay can never read bytes the proof did not vouch for. The 16 MiB bound survives as a per-record framing limit, which is the only remaining way the source could become resident. claude-transcript-branch-proof.ts is split at its real seam to stay under max-lines: claude-transcript-branch-graph.ts is what the rows mean as a graph, and the proof file is which bytes the graph gets to see. Peak heap over a 252 MiB transcript: 542 MiB whole-file, 53 MiB streaming — and flat at 53 MiB for a 63 MiB transcript, where whole-file took 136 MiB. * fix(claude): fail closed when the ancestry walk misses the anchor The source-budget anchor test filtered on `"latest"`, which also removed the last-prompt marker. The transcript was unprovable, so the empty window and single pass it asserted came from an INCONSISTENT verdict, not from the leaf-equals-anchor path. Filter the record only and assert the boundary. `ancestryChain` returned [] both for "the leaf IS the anchor" and for a walk that fell off the graph. The first means nothing followed the anchor; the second means we never looked. Throw on the second, so the window reports an inconsistent boundary rather than non-delivery. |
||
|
Neil
|
253f0e3946 |
Fix Antigravity source-control model discovery and retired defaults (#21606)
* fix(antigravity): discover current source-control models and use CLI defaults * fix(antigravity): gate configured models on remote runtime support * fix(runtime): forward default TUI agent for remote git generation * test(runtime): cover inherited agent forwarding |
||
|
Jinjing
|
00da5fd556 |
test(worktrees): add comprehensive nested lineage coverage (#21903)
- Add 10 test cases for nested worktree rendering and collapse behavior - Handle edge cases: cycles, uneven siblings, multiple depth levels - Extract stopNestedWorktreeCardBubble to shared header-event-guards module |
||
|
SahilZ0810
|
ffb79c71e0 |
fix(editor): open plain details blocks in rich markdown mode (#19784)
* fix(editor): allow plain details blocks in rich markdown mode * fix(editor): preserve case-sensitive details class values |
||
|
Jinwoo Hong
|
2d697a4012 |
test(config): count the haptics notify module in the session route's page closure (#21908)
#21864 (haptics on the page) and #21871 (mermaid on the page) were each green against a main that lacked the other. Together, `haptics.web.ts` reaches `bridge-haptics-notify.ts` inside the session route's closure, so the module pin recorded by #21871 reads 4324 on main, not 4323. Pin the measured count and name the module. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
|
Neil
|
27a0889dcf | test(relay): account for OpenCode marker in OMP launch environment (#21907) | ||
|
Neil
|
35005fb65c |
fix(pi): keep panes working while async subagents run (#21882)
* fix(pi): wait for async subagents before settling pane * fix(pi): handle subagent event aliases and reloads * test(pi): assert lifecycle listener cardinality |
||
|
Jinwoo Hong
|
e9b180685b |
feat(mobile): render Mermaid diagrams on the page from one deferred engine artifact (OTA phase C, C7.10 B) (#21871)
* test(mobile): measure mermaid rendered in the page Red-first for C7.10 item B. The check mounts the real web sibling in chromium and webkit under the shipped shell CSP and asks four things of it: that a diagram renders with zero policy violations and zero eval / new Function calls, that the SVG is the native buildHtml's own output once the diagram id and xmlns:xlink are normalised away, that a hostile diagram lands inert, and that a source change, an unmount and a remount leave exactly one SVG and no listener of the first mount. The equality oracle is buildHtml itself, bundled for Node behind a Proxy stub for its native imports and served as its own document in the same browser, so neither side of the comparison is retyped. All eight cases fail on this commit: the sibling is still the labelled source box. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): fence the session download rather than its module list Ruling 28. mobileWebAppRouteClosure reads metafile.inputs, which holds dynamically imported modules under splitting: true exactly as it does under splitting: false, so it cannot say "on demand" about anything: an on-demand mermaid moves the session route's module list 4320 -> 6362 while its download does not move at all. So the fence moves to entryStaticClosure. The new helper walks the emitted chunks from the output the route's own module landed in and follows import-statement edges only, and hands back both halves, because mermaid's absence from the download is only a measurement while its 66 files are present in the deferred half. The module list's new total is recorded in the docstring with its reason and asserted beside the engine's own file count, which moves only when the pinned mermaid version does. Red on this commit: no mermaid in the closure yet. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * feat(mobile): render mermaid in the page The web sibling stops being a source box. mermaid is a browser library, so the page imports it inside the render effect and draws the diagram in this document: no WebView, no 3.7 MB engine string, and nothing of the engine downloaded by a session with no diagram on it. What replaces the sandbox is mermaid's own securityLevel: 'strict', which runs its serialized SVG through DOMPurify. The native path's </script> escaping has no analogue here and needs none, because the source is a JS string argument rather than text spliced into an inline script. Measured in both engines: a script in a label, a </script>, an onerror and a javascript: click all land inert. The configuration is now one object both hosts read, so the theme cannot drift between the page and the phone; buildHtml serializes it instead of holding a second copy. It gains suppressErrorRendering, because mermaid otherwise draws its own error diagram into a temporary element and leaves that element behind when it rethrows -- an orphan SVG on the page, and on native a diagram the component is about to replace with the source box anyway. The dispose clears the host on unmount and on a source change; the id is a useId, because mermaid writes it into the stylesheet inside the SVG and it has to be a CSS identifier. Also re-records the closure total the previous commit pinned: with the real component the session route's module list is 6376, not the design probe's 6362, and the reason is in that file's docstring. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): budget the deferred engine's chunks apart from the routes Putting mermaid on the page took the app bundle from 69 emitted scripts to 172, and the asset budget failed: 215 assets against a ceiling of 115. The cause is not a page split running away, which is what that ceiling is for -- it is that mermaid lazily imports each of its own diagram types, so one import() lands 103 scripts no route count predicts. So the ceiling gains a second term, named and measured (172 scripts with mermaid against 69 with it aliased to a stub, at 11.17.2), rather than the route term being raised to cover it. A page split running away still fails on the route term, and the failure still says which of the two grew. The consequence is worth reading twice: the derived ceiling has to stay inside the 256 assets the shell will load, and with 42 images it now crosses that at 24 routes instead of 50. The bundle is at 215 today with 14 routes, so there is room for about ten more routes before a green build produces a manifest no phone will open. Measured by the config/scripts suite failing on this head, not predicted. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * refactor(mobile): pre-bundle the page's mermaid into one artifact import('mermaid') from inside the app bundle emitted 103 scripts, not one: mermaid lazily imports each of its own diagram types and esbuild splits along those boundaries. Every one of those scripts sits inside the OTA generation the phone has already downloaded, so the split moved no bytes over the wire and spent 103 of the 256 manifest assets the shell will load -- which is the scarce resource here, and the reason the previous commit had to invent a second ceiling term. So a sibling generator bundles the package into one ESM module beside the WebView engine it already builds, emitted by the same postinstall run, gitignored and lint-ignored with the others. The page imports that artifact on demand instead, through a loader whose return type names the two calls the component makes -- checked against the artifact's own inferred export rather than cast to it. Measured, at 14 routes: emitted scripts 172 -> 69 (68 with no deferred engine at all) manifest assets 215 -> 112 (111 with none) session modules 6376 -> 4323 (+3 over main: config, loader, artifact) chunks fetched for one graph TD 27 -> 1 bytes fetched 837,530 -> 3,482,965 The static-closure fence is unchanged in meaning and now reads on the artifact: absent from every chunk the route reaches by an import statement, present in the deferred half. The rendered SVG is byte-for- byte what it was, so the equality against the native document still holds on both engines. Also adds the diagram to the webview-consumers list, which is what that list means: its native component imports the package and its sibling is what the builder resolves instead. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * revert(mobile): drop the deferred-engine ceiling term, keep the control With the engine pre-bundled into one artifact the bundle emits 69 scripts at 14 routes against the route term's 72, so the second term this series added has nothing left to do and the route count is the only term again. mobileWebAppBundleMaxChunks and the asset ceiling derived from it are back to what main has; the shell's 256 assets are crossed at 50 routes again rather than at 24. What stays is why. A ceiling raised to admit 172 scripts would have admitted any split at all, so the budget test gains the control that holds the line: the single-artifact count passes the ceiling and the lazily-chunked count fails it, both measured at 14 routes, with mermaid named as what produced the second. Red before the term came out: the control failed asserting 172 > 175. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): keep build output out of the raw-request-port census The census walks mobile/src for AST reaches into the unvalidated request port, and the pre-bundled mermaid artifact is the first generated file under src that is executable code rather than a string literal. Two of its own vendored dependencies contain the token `sendRequest`, so the walk read minified third-party code as a new call site and asked for an inventory line nobody can ever migrate. So `*.generated.ts` joins node_modules and test files in that file's stated list of what it does not scan, with the reason. The scripts that emit those artifacts are ordinary source and are still scanned, which is where a real reach would be. Two halves to the new control, because a filter that skipped everything would satisfy either alone: nothing generated is left in the scan, and the matcher still finds the port when handed one line of code. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): escape the shared config into the native inline script buildHtml spliced JSON.stringify(MERMAID_DIAGRAM_CONFIG) straight into the inline <script>, twenty lines below the function that exists because JSON.stringify leaves `<`, `>`, `&` and the U+2028/9 separators raw. Inert at today's five hex colours, and not inert for a themeCSS or a font stack, which is free text going into the same script element. So the escaping splits from the stringify and both callers use it: the source keeps its own wrapper, the config gets one. Those characters only ever appear inside JSON string literals, so escaping them is valid for an object serialization exactly as it is for a string. Red first: a config carrying `</script><script>` put four raw closers in the document where a benign build has two. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): pin the page mermaid type against the package's own The loader returned the artifact's default as PageMermaid, which checked that two names exist and nothing about their shapes: the artifact is minified vendor output and both members infer as `any` there -- a probe assigning engine.render to a number compiles -- and `any` satisfies every signature there is. So the shapes are asserted against the package's `Mermaid`, which is precise. A PageMermaid member whose signature the engine does not really have now fails at this line rather than at a call the page makes. In the product module, not a test: mobile/tsconfig.json excludes test files, so a type-only assertion in one is never compiled. Underscored because it is a compile-time statement with no runtime reader, which is the form the linter asks for. Control, verified both ways: changing render to (id: number) => Promise<{ svg: number }> reds tsc naming both parameter and return, and the real signatures compile. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * docs(mobile): re-measure the chunk series and say what it does not show The four-point series was stale and read as a slope it is not. Measured again on this head, by copying the route tree and dropping routes from the end of the sorted key list -- both siblings of each, because deleting a .web.tsx alone leaves the native file for the builder to resolve and measures an entirely different closure, which is how the first attempt at this produced 77 scripts for 14 routes: 8 routes -> 32 scripts 10 routes -> 43 12 routes -> 61 14 routes -> 69 (the real tree) Between four and nine more per route depending on which route, so 4r + 16 is a bound and not a fit, and the justification now says that instead of claiming three per route. It also says the part that matters more: at 14 routes the tree measures 69 against 72, and the last two routes cost the 8 the ceiling grants for two. The fence is at break-even, and the new assertion states that slope from the function rather than from a comment. Also records what the generation weighs, since every chunk ships in it whether or not a phone fetches one: 8,016,714 bytes across 112 assets against the 9 MiB ceiling, 84.9%, 1,420,470 left. It was 4,539,090 before item B, and the engine is the difference. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): pin the native fallback under suppressErrorRendering The shared config reaches the phone too, and it gained a key the native path did not have. So the native document is now loaded for a diagram that throws, in both engines, with window.ReactNativeWebView standing in for the host: mermaid's run still rethrows, the document's own catch still posts `error`, and that is the message the component turns into the source box. Measured both ways, so the case says which half the key owns. Whether the fallback fires does not depend on it -- `error` is posted with the key and without it. What depends on it is that nothing is drawn behind the fallback: removing the key leaves mermaid's own error diagram in the document and reds this case at 1 SVG against 0, on chromium and webkit alike. The control is the same document for a diagram that parses: a height, not `error`, and one SVG. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * refactor(mobile): one walk for every source census, without build output Nine censuses under mobile/src each held a copy of the same recursive walk, and each decided for itself what a source file is: seven had no opinion about generated files, one excluded them in its own regex, and one had the exclusion I added last round. So all nine read 7.9 MB of emitted vendor code -- 3.7 MB of mermaid for the WebView, 3.5 MB of it for the page -- and the two largest censuses TypeScript-parsed all of it, looking for call sites nobody wrote and nobody can move. That is what took rpc-params-contract-type-only-boundary over its 5 s timeout in CI once the fifth artifact arrived. Measured here, median of 3, import plus tests: main, 4 artifacts, no exclusion 1004 ms (slowest case 831 ms) with the 5th, no exclusion 1513 ms (slowest case 1358 ms) with the 5th, this commit 947 ms (slowest case 788 ms) So it lands below where main has it, not merely below where I left it. Across the nine, four more halve: rpc-operation-cast-fence 769 -> 441, rpc-subscription-boundary 946 -> 468, unchecked-rpc-reader-boundary 1042 -> 538, lifecycle-owner 747 -> 433, reanimated-web-mapper-deps 1028 -> 516. The two that already excluded generated files do not move. What each census counts as interesting -- extensions, whether test files are in -- stays its own, because they genuinely disagree. What counts as a source file at all is now said once. The control is the file that started it: a *.generated.ts whose text holds exactly the import a census is hunting, planted beside an ordinary file carrying the same text. The generated one is not returned and the ordinary one is, so the absence is a measurement. A second control reads mobile/.gitignore and holds the predicate to every artifact the tree generates, and a third fences the walk itself to one spelling, so a tenth census cannot paste the cost back in. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * docs(mobile): correct why the type pin sits in the product module The comment said a type-only pin in a test file "is never compiled". That is false: mobile/tsconfig.json excludes *.test.ts, but tsconfig.test.json is a second program that does check them, run by check:tests-typecheck and held by the tests-typecheck ratchet. The conclusion is unchanged and the reason is now the true one. The app's own typecheck is the unconditional gate and would not cover a pin written in a test; the test program is real but carries a grandfathered baseline and a few files held outside it on purpose. And the assertion is about this module's own type either way, so it belongs beside it. Comment only; tsc, the ratchet and both lints re-run on the file. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
|
Jinwoo Hong
|
c77a82f783 |
feat(mobile): fire the page's haptics over the bridge notify (OTA phase C, C7.10 E) (#21864)
* feat(mobile): give the page a haptics notify and the grant that gates it `native.haptics.trigger` joins the envelope's notify union with a `kind` of exactly the five `src/platform/haptics.ts` has, and the single token `haptics` joins `BRIDGE_NOTIFY_GRANTS` and `MOBILE_WEB_SHELL_GRANTS`. A notify rather than a verb because nothing is owed back: a reply would spend a slot in the same 64-deep in-flight window a forwarded request does, and there are 90 call sites in this app, some of them one per row of a scrolling list (rulings-ota-c7.md ruling 30). The arm's fields live in their own module because `bridge-envelope.ts` is at its line cap, as `bridge-event-envelope-bytes.ts` already is; the version literal stays in the envelope, so the fields are spread in beside it rather than reading it back through an import cycle. The shell's half rides `onHaptic` on `BridgeHostOptions`, as every other device-local notify does: the host is the protocol's side of the bridge and a static import of the app's haptics would put `react-native` and `expo-haptics` in its graph, which breaks every test that loads it. `page-haptics.ts` is the one mapping — `haptics.ts`'s own functions, its `Platform.OS` split and its Android `HapticFeedbackConstants` untouched. The dispatch branch rides along with the union rather than waiting for the page side: `Record<BridgeNotifyName, …>` and the `notify` fall-through are total over that union, so the shell does not compile without it. That is the totality working, and `bridge-notify-grants.test.ts` shows it as the TS2741 a missing row is. Red first: the envelope cases per kind, the ungranted refusal, the grant-list pin and the missing-row type error all failed against the tree before this. Control on the dispatch: neutering `options.onHaptic` reds 2 of the 29 cases. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * feat(mobile): post the page's haptics over the notify instead of doing nothing `haptics.web.ts` stops being five no-ops. Each of the five posts its own kind through the notify seam the entry publishes — the same shape `publishExternalLinkOpener` has, and for the same reason: every caller is a plain function inside a row's press handler that no provider wraps. `notifyHaptics` joins the page client beside the other gated notifies and answers whether the frame left, which nothing reads: a tap that did not buzz is what the page did before this, and a warning per refusal would be one per row of a scrolling list. Measured off the frame the client posted rather than a written copy of its shape, which is what drifts: 77 / 74 / 72 / 70 / 73 bytes for mediumImpact / selection / success / error / edgeBump, the widest under 0.012% of `BRIDGE_MAX_MESSAGE_BYTES`, and a twelve-row scroll 888 bytes across twelve frames. The `web-overrides.json` reason now says what the file does instead of what it declines to do. Red first: the nine web-seam cases failed on `publishHapticsNotifier is not a function`, and the six client cases on `notifyHaptics is not a function`. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * feat(mobile): grant haptics on every page route, with a census that derives the list All five declared page routes carry the `haptics` grant, and the list is a measurement rather than a hand choice: `WorktreeListRow` is in every page closure and calls the seam, so a route without the grant is a page whose taps stop buzzing with nothing on screen to say why. Grants are resolved once from the route the shell opened and held for the session, so the declaration is the only place to fix it. `mobile-web-app-haptics-seam.mjs` is the shared walk, beside the external-link one: it reads the kinds off the tuple that declares them, finds every exported `trigger…` function in a haptics module, and reports the kind each one posts. The posting call is found through the binding `publishHapticsNotifier` assigns rather than a local spelled `post`, because a rename would otherwise turn every posting site into a non-posting one and leave this green on a page with no haptics at all. The census proper holds each route's closure to the `.web.ts` sibling, asserts at least one importer so the grant is not idle, and derives the granted-route list from the closures. The control is the design's: the same walk over the native sibling finds the same five functions and no posting site, so "all five post" is a number rather than an empty scan. Controls run: dropping `haptics` from one route reds 1 of 23; neutering one web post reds 1 of 23. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): record what the haptics notify costs a page closure One module. Every page closure grew by exactly `bridge-haptics-notify.ts`, and it arrives through `page-route-policy.ts` reading the grant token rather than through the seam, whose import of the kind type is erased; its only dependency is `zod`, which the envelope already put in every closure, so the module total moved by the same one. Local counts per route went 294 → 295, 379 → 380, 435 → 436, 309 → 310, 335 → 336. Pinned structurally rather than as a total, because an absolute closure count is main's to move and a number that drifts for unrelated reasons is one nobody reads. The call sites this replaces, measured over product modules: `triggerError` 43, `triggerSuccess` 24, `triggerSelection` 12, `triggerMediumImpact` 10, `triggerEdgeBump` 1 — 90 across 35 importing modules, which is the design's count plus `page-haptics.ts` itself. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): carry the haptics grant into the shell's two grant pins `bridge-host-init.test.ts` names the grants `init` issues, so the token belongs in that list. `MobileWebShellScreen.test.tsx` now mocks `expo-haptics` for the reason it already mocks the clipboard and both pickers: the screen hands `playPageHaptic` over and reaching the real module pulls in an Expo runtime this test does not have, which failed the whole suite at import. Which expo member each kind reaches stays in `page-haptics.test.ts`. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * refactor(mobile): map each haptic kind to a named import, not a namespace index The changed-code gate refuses a computed reference into an imported namespace, in both the mapping and its test, and it is right to: `haptics[NAME_BY_KIND[kind]]()` is a call nothing can follow. Each function is a named import instead, which also keeps the second compile-time direction — a row naming something `haptics.ts` does not export is now an import error rather than a `keyof` mismatch. The third direction moves with it, from a namespace read in the test to the census that already reads both files' text: `hapticsImportedNames` names what the shell's mapping takes from the app's haptics, and the census holds that to the five the native file exports. So a haptic added there with no kind of its own still fails, and now it fails where the other two siblings' names are already compared. The test's two `as` assertions become one annotated hoisted type, the shape `MobileWebShellScreen.test.tsx` uses. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * docs(mobile): state the true reason the haptics grant is one token `GRANT_NAME_PATTERN` accepts `native.haptics.trigger` — it admits `native.<a>.<b>` with lowercase segments, which is why it rejected `native.media.readChunk` and rejects `navigate-back`, not a dotted name as such. So four comments claiming a route declaring the notify's own name would have its bundle refused were false, and they are gone: the grant is a token because the notify table's grants are tokens, a notify not being a verb, and the dotted names in `MOBILE_WEB_SHELL_GRANTS` are spread from the verb table alone. Also folded, with the false claim: `implementedPageRoutes` filters on `grants.every(implementsGrant)`, so a token every page route declares couples the whole set to a shell that carries it — against one without it, no page route is served at all and the phone renders five native screens. Stated in the function's docstring and beside the census's derived list, and pinned: the same declaration under a grant this build does not implement comes back empty, with the token-free route as the control. Removing `BRIDGE_HAPTICS_GRANT` from `MOBILE_WEB_SHELL_GRANTS` reds that case. `%#` consumes no argument, so the web seam's five cases were titled with the whole function body; the kind is the first element now and `%s` names it. One 110-char comment line in `bridge-client-notifications.ts` wrapped to the file's 100; the two still over it there are main's. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
|
Jinwoo Hong
|
ba7583244b |
fix(editor): persist PDF zoom across tabs and restarts (#21879)
* fix(editor): persist PDF zoom preferences * fix(pdf): avoid path-only zoom persistence |
||
|
Neil
|
7b97551acf |
fix(opencode): isolate v1/v2 plugins and preserve WSL config (#21900)
* fix(opencode): include cache read and write usage totals * fix(opencode): satisfy aggregate query safety checks * chore(i18n): refresh runtime English catalog * fix(opencode): isolate plugin variants and preserve WSL config |
||
|
Jinwoo Hong
|
8a3a119052 |
fix(i18n): regenerate the runtime catalog and localize the cookie example from #21462 (#21889)
* fix(i18n): regenerate the runtime-required English catalog for the AccountsPane strings
#21462 (
|
||
|
Neil
|
87a22db25a |
fix(opencode): include cache read and write usage totals (#21886)
* fix(opencode): include cache read and write usage totals * fix(opencode): satisfy aggregate query safety checks * chore(i18n): refresh runtime English catalog |
||
|
Jinwoo Hong
|
169bccc544 |
fix(mobile): restore query-string's named exports under the 9.5.1 override (#21727)
#21652 overrode query-string to 9.5.1 for GHSA-vcc3-ghjq-m6fr (decode-uri-component <= 0.4.2). 9.x's entry exports only `default`, while expo-router's linking forks, @react-navigation/core and @react-navigation/native all `import * as queryString`, so `stringify` and `parse` became undefined: any push carrying a param outside the path pattern and any href with a query threw. Patch the entry to re-export the named API; the override and the advisory fix stay. The lockfile carries the patch hash only; regenerated by hand because a non-frozen install re-resolves peer suffixes across the file. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
|
Jinwoo Hong
|
58a80d996b |
test(runtime): expect the agent's own submit delay in the PTY timing policy case (#21874)
#21665 gave antigravity a per-line settle before Enter, so the delay the test computed from bytes alone is 45 ms short of what the runtime waits. Under fake timers that leaves the submit pending until the real 30 s timeout, which is what every PR's node shard 8/8 has been failing on since it landed. The case now derives the expected delay from the agent's policy, so a future per-agent term moves the expectation with it. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb |
||
 Wooseong Kim
|
7a6d10064e |
fix: read OpenCode Go usage from the console API (#21462)
* fix: read OpenCode Go usage from the console API The workspace HTML page now 302s to console login. Fetch /console/api/go/status with x-org-id, map JSON meters into the existing windows, and keep __Host-console_session on the closed cookie allowlist. Fixes #21420 * fix: tell users to paste the OpenCode console session cookie The Go status API is authed by __Host-console_session. Settings still told people to paste auth only, which 401s. Ask for the full Cookie header; auth remains enough for workspace discovery. |
||
  Pablo Werlangandorca-agent
|
646fa3645f |
fix(opencode): attribute shared-server sessions to their panes (#21577)
* docs: allow-list opencode tool-readout follow-up note * fix(opencode): attribute shared-server sessions to their panes The v2 shared server stamps every hook post with its own frozen pane, so all panes' status lands on the starter pane (#21359). - shared: session->pane registry plus ingest-time envelope rewrite; bound sessions resolve to their real pane, tab and live launch token before disposition, unbound sessions keep the stamped identity. - main: binder poll (SQLite session store, PTY-registry pane snapshots, argv-aware client sweep) with directory-containment plus client-lifetime correlation; 60s loop plus debounced SessionStart kick, wired into the hook server lifecycle. * fix(opencode): newest-wins pane dedupe, macOS private/tmp normalization Live verification against the dev instance found two binder gaps: remint rows for one pane counted as an ambiguous tie, and /tmp vs /private/tmp spellings never met on macOS. * fix(opencode): review fixes — newest-wins worktree, drop dead constant - applyBinderOwnerships now overwrites per-pane worktree, matching the round's newest-wins pane dedupe; a remint's live row wins over a stale row (pinned by test). - remove the unused OPENCODE_CLIENT_PRE_CREATE_WINDOW_MS export and the nowMs residue from clientCouldCreate. - give the per-pane launch-token cache its own named cap constant. * fix(opencode): address thread review — cursor, native table, tokens, lifecycle - composite (time_created, id) store cursor advanced past handled rows only, so same-millisecond pagination and full unbound maps no longer drop sessions silently. - Windows sweep reads the native process table instead of forking powershell.exe; quote-aware argv parsing on both platforms. - directory keys via normalizeRuntimePathForComparison (Windows case-fold, POSIX backslash literals) plus narrow macOS /tmp|/var|/etc aliases and lexical dot-segment resolution. - bound sessions always take the stored pane token (never the frozen stamp); token tracking runs after resolution. - binder generation guard discards post-stop rounds; first round runs immediately at loop start. - unbind/move use exact pane-key match; pane launch-token cache gets its own cap constant. - move the tool-readout note out of this PR for its own branch. * fix(opencode): second review round — executable field, worktree scope, round lifecycle - POSIX sweep reads comm= alongside args= and classifies on the kernel executable name, so unquoted install paths with spaces no longer split argv[0] and reject the client; Windows rows carry the native table name. Degrades to argv[0] when comm is unavailable. - bound sessions take only the binding's worktree (never the stamped pane's), so a worktree-less binding cannot file a row under the wrong worktree. - the binder generation is captured before the round body and the running flag clears only for the current generation, so an obsolete post-stop round cannot admit an overlapping round. --------- Co-authored-by: orca-agent <orca-agent@local> |
||
|
Jinwoo Hong
|
e476193bf5 |
chore(relay): bound the shadow health gate and apply a pending backend update on resume (#21865)
* fix(relay): bound the same-cap shadow gate and apply a resumed backend update Two findings both adversarial reviews of tonight's merged set agree on. The report-only shadow health gate (#21849) had `continue-on-error: true` but no step timeout. That bounds the step's contribution to the job outcome, not its clock. Its reads are serialised, and a failure that answers nothing slowly — an expired credential, a project-wide Logging 429 storm — makes every read cost its full 3 x 60 s retry budget, so the cost scales with the roll window: roughly 8S + 2 reads for S ten-minute sub-windows. A 40-minute window is about 34 reads, or 108 minutes, against the job's `timeout-minutes: 75`. A cancelled job cannot be absorbed by continue-on-error, fires the failure-gated cleanup isolation on an already-restored cell, and stops the strict next-cell chain. Give the step `timeout-minutes: 5` and the artifact upload `timeout-minutes: 2`. A timed-out step is a failed step, which continue-on-error covers, so the job stays green. Inside the script, stop reading after an overall four-minute deadline and report the remaining checks unverified, so the normal outcome is a written verdict rather than a killed process; the step timeout is then only for a hung process. The census test pins both timeouts and that the deadline leaves the step time to write its verdict. The resume branch (#21860) accepted `changes == 0` with a non-empty `backendUpdate` as complete and applied nothing, so a resumed cell silently kept the 300-second drain and no request logging behind a green resume. That shape means the template and MIG are converged and only this cell's reviewed backend update is left, so apply the saved resume plan — the validator has already bounded it to this cell's backend and neither attribute restarts an instance — then continue as converged. Template-and-MIG drift still applies nothing, which is what a resume means, and a stranded cell's explicit MIG replace is unchanged. Claude-Session: relay-same-cap-gate-timeout-and-resume * fix(relay): raise the shadow gate bounds clear of a healthy gate's read time A healthy gate is already minutes of serial reads on the 2-vcpu runner, so a four-minute deadline would report unverified tails on ordinary days and stop the shadow roll measuring the comparison it exists for. Raise both together: the step to eight minutes and the script's own deadline to seven, keeping the census pin that the deadline leaves the step room to write its verdict. The job budget is unaffected: a ~14-minute cell plus eight is well inside 75. Claude-Session: relay-same-cap-gate-timeout-and-resume |
||
|
Wooseong KimandNeil
|
30f2bc60f9 |
fix(antigravity): recognize non-Gemini tui-idle prompts (#21231)
* fix(antigravity): recognize non-Gemini tui-idle prompts * fix(antigravity): reject stale composer caret in model picker * fix(antigravity): do not treat a wrap continuation caret as ready An unsent composer can show `> draft` then an indented `>`. That continuation is not an empty input box, so tui-idle must stay false. * test(antigravity): align later bare-caret status expectation --------- Co-authored-by: Neil <neil@stably.ai> |
||
  Seongho BaeandCursor
|
ea5152f1c2 |
fix(orchestration): line-settle delay for antigravity multiline paste (#21665)
* fix(orchestration): retry Enter after cursor-agent worker-start paste Worker-start dispatches through bracketed paste in the main process; cursor-agent can leave long prompts as "Pasted text +N lines" and swallow the first Enter. Apply the same submitRetryDelayMs path Codex uses in the renderer, but only for agents without the Claude/Codex render gate so hook turn-start reservation stays intact. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(orchestration): line-settle delay for antigravity multiline paste Antigravity 1.2.x expands long bracketed paste slowly ("↑ N more lines") while Orca only waited for byte ingest (~500 ms on macOS). Add submitLineSettleMsPerLine and retry Enter for antigravity; wire agent-aware submit scheduling through the main-process prompt writer and plain terminal.send suffix path. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(orchestration): antigravity line-settle only; drop unverified retry Address PR review: revert accidental pnpm-lock.yaml churn, remove cursor and antigravity submitRetryDelayMs until live-verified, keep submitLineSettleMsPerLine for agy multiline paste, and move the regression test out of the 900+ line runtime submission suite. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
Neil
|
bd860f34c1 |
fix(mobile): normalize OMP terminal momentum across refresh rates (#21687)
Use elapsed animation-frame time for OMP terminal momentum and update the generated payload contract. |
||
|
Neil
|
438744ca77 | fix(opencode): preserve global config discovery (#21854) | ||
|
Jinwoo Hong
|
2524737ef0 |
chore(relay): apply the cell backend drain and request-logging settings inside each same-cap wave (#21860)
* chore(relay): target each cell's backend service from the same-cap job
The 60 s connection drain timeout merged in #21848 has no safe apply path.
A root plan scoped to the backend services alone still pulls every
`google_compute_instance_template.relay_gce_cell` in as a dependency, and
standing image drift turns all 29 into replacements, so applying it would roll
the fleet at once.
Add `google_compute_backend_service.relay_gce_cell["${TARGET_CELL_ID}"]` to
both plan invocations in the per-cell same-cap job, next to the template and
MIG it already targets, and teach the reviewed plan validator to allow exactly
one extra change: an in-place update of that one cell's backend whose only
changed attribute is `connection_draining_timeout_sec`, landing on the
constant `validate-relay-asia-topology-plan.mjs` exports. Any other attribute,
any other resource, or a backend for another cell still fails the validator.
The accepted update is reported as `connectionDrainUpdate` and kept out of
`changes`, so the apply step's stranded branch and the resume step's drift
branch keep reading the template-and-MIG count they were written against; the
resume branch additionally accepts a plan whose only pending change is that
drain update, which restarts nothing.
Claude-Session: relay-same-cap-targets-cell-backend
* fix(relay): also let the same-cap wave apply this cell's LB request logging
A read-only production plan for production-gce-c7 showed the live US cell
backends carry no `log_config` at all, while relay-gce-cells.tf has declared
`log_config { enable = true, sample_rate = var.relay_gce_cell_log_sample_rate }`
on every cell backend since the Terraform root landed in
|
||
  buf0-bot[bot]andbench
|
41f34f6ff2 |
fix(terminal): let Shift+middle-click paste in mouse-tracking panes (#21858)
* fix(terminal): let Shift+middle-click paste in mouse-tracking panes Follow-up to #21834 (issue #21762). That fix arms the native-paste suppression window for every terminal middle-click, then returns early when the pane is in mouse-tracking mode so the TUI performs the paste from the forwarded mouse report. xterm's SelectionService.shouldForceSelection deliberately withholds that report for a shifted click (Option-click on Mac), so the TUI never pastes. With the native follow-up paste now suppressed as well, Shift+middle-click in Claude Code, Codex, and other tracking TUIs pasted nothing at all. Previously Chromium's native paste was the one paste. Mirror xterm's platform rule: when the click's modifier forces selection, fall through to Orca's own paste-to-PTY path (stop propagation, focus, paste) exactly as in a non-tracking pane. The auxclick handler gates stopPropagation the same way. * test(terminal): pin Alt+middle-click to the TUI-owned path off Mac --------- Co-authored-by: bench <bench@example.invalid> |
||
|
Neil
|
e0b717dd60 | test(antigravity): cover Kitty mode reattach metadata (#21810) | ||
|
Neil
|
76d87604ff |
fix(terminal): keep drag selection stable during redraws
Pause output-driven drift fitting while the user drags a terminal selection, then converge immediately on mouseup. Includes regression coverage and visual proof. |
||
|
Neil
|
3b055c869f |
fix(wsl): await Pi and OMP guest relay materialization (#21721)
* fix(wsl): await Pi and OMP guest relay materialization * fix(wsl): materialize Pi extension before guest launch * fix(wsl): keep relay state under lint limit * fix(wsl): preserve guest agent readiness across launches * test(wsl): expect guest status path translation * ci: rerun PR checks after rebase * ci: retrigger PR checks * ci: run final PR verification |
||
|
Jinwoo Hong
|
5b8ac36f41 |
chore(relay): add a report-only post-wave health gate to the same-cap cell job (#21849)
* feat(relay): report a post-wave health verdict on each same-cap cell, without gating on it After a same-cap cell finishes rolling, an operator reads five things by hand before dispatching the next cell: director 503s against the same clock hour a day and two days earlier, whether the cell's new container announced its listener and has stayed up, the cell's own pool pressure, the asia-east2 pool trio, and Cloud SQL FATALs. This runs those same reads automatically and records PASS / WARN / WOULD_BLOCK with its numbers, so its calls can be compared with the operator's over a full roll before it is ever allowed to stop one. It cannot fail a cell in this change. The script exits 0 on every verdict, and the step is continue-on-error, so even a crash stays off the job's outcome and the failure failsafe cannot fire on anything it observes. It also runs after the restore, so no cell waits on it to go back into admission. Cloud Logging returns only --limit entries and says nothing when it truncates, so every count is split into sub-windows of ten minutes and a sub-window that comes back at the limit is reported unverified rather than as a count. Windows are always explicitly bounded: --freshness does not bind on these logs. Claude-Session: https://claude.ai/session/ced32ebb-7155-4413-adad-1eccd14c2010 * fix(relay): bound the shadow gate's cell reads at the apply start and cap every read Four fixes from review, all in the report-only shadow health gate. The boot search opened at apply-completed-at, which is stamped after `terraform apply` and `wait-until --stable`. The new container announces its listener while the MIG is still converging, so that bound is already past the announcement it looks for and a healthy roll read as would-block. The job now stamps apply-started-at immediately before the apply, and the boot search opens there; apply-completed-at is kept, recorded rather than judged, so an operator comparing verdicts can see apply time next to boot time. The crash query started at the newest listener timestamp, which erased any crash before it. A crash-restart loop ends with an announcement that looks like a clean boot, so that is exactly the case it hid: against production, the 2026-09-20 c28 crash at 20:18:10 was dropped because the listener landed at 20:18:27. It now runs from the apply start, still scoped to the instance id the listener identified, and that crash is counted. A runtime-metrics read that came back at its 500-entry limit fed judgePool as though it were a complete sample run. A truncated run has holes and the consecutive-sample rule reads a hole as a recovery, so it now reports unverified. gcloud reads had no timeout. continue-on-error bounds the job's outcome but not its clock, so a stalled read could have spent the rollout's remaining minutes. Each read now gets 60 s and a timed-out read is just a failed read. Claude-Session: https://claude.ai/session/ced32ebb-7155-4413-adad-1eccd14c2010 * test(relay): require each shadow-gate stamp's presence before asserting its order The ordering assertion used indexOf, which answers -1 for an absent stamp, and -1 precedes every real offset. Deleting the apply-started-at line left the test green, so the census could not see the fix it was written to pin. Each stamp's presence is now asserted first, with a message naming the stamp and the step, and presence is judged inside the step that owns the stamp rather than anywhere in the file: a stamp written into a neighbouring step records the wrong instant but would satisfy a whole-file match. Control-run against a scratch copy of the job. Deleting drain-started-at, apply-started-at, or apply-completed-at each reds with its own message, and moving apply-started-at after terraform apply reds on the ordering assertion, so presence and order both fail independently. Claude-Session: https://claude.ai/session/ced32ebb-7155-4413-adad-1eccd14c2010 |
||
|
Jinwoo Hong
|
4f839cc8c9 |
chore(relay): cut the cell LB connection drain to 60 s and allow ten-cell same-cap batches (#21848)
* perf(relay): cut the cell LB drain to 60s and widen the same-cap batch to ten cells Two independent sources of relay roll wall clock, neither of which protects a host: 1. `connection_draining_timeout_sec` on the per-cell backend services was 300s. The same-cap job drains every host off the cell to a restart-safe condition before Terraform runs, so the LB drain only ever covers a host still mid-handshake. Measured 2026-09-16 over ten same-cap cell jobs, it sat as ~5m55s of dead time between `Apply complete` and the old VM powering off, inside an 8.5-minute `wait-until --stable` step. Now 60s, and pinned in the topology `check` block beside the other fixed-one invariants. 2. The same-cap wave capped a batch at four cells, so a 22-cell roll needed six batches, six single-use monitor gates, and a human handoff per batch. The wave workflow now declares cell_1..cell_10 with the identical serial shape and chaining, and the validator accepts two to ten. The shared wave-index rule (`relay-monitor-evidence.mjs` and the relay-ops preflight CLI) widens from 0-3 to 0-9 so the later cells can present the same evidence; each job workflow keeps its own narrower range, so the capacity wave stays at four. Cells remain strictly serial, one at a time behind the rollout lease, each with its own live preflight. Claude-Session: https://claude.ai/session/relay-roll-drain-timeout-and-batch-cap * fix(relay): align the Asia topology plan validator with the 60s cell drain `validate-relay-asia-topology-plan.mjs` rejected any Asia backend whose `connection_draining_timeout_sec` was not 300, and `cloud-deploy-relay-asia-topology.yml` targets `google_compute_backend_service.relay_gce_cell["<cell>"]` per cell. With the Terraform local at 60 that workflow would have failed its own plan review. The validator's two restated topology values are now named exports, and a new census test reads `relay-gce-cells.tf` and equates three statements of each: the `relay_gce_topology` local, the topology `check` assert that pins it, and the validator constant. Terraform cannot export a local to JS, so reading the source is the only way to stop them drifting; the test was confirmed to fail when the local alone is moved back to 300. Repo-wide grep finds no other pin of the drain value. Claude-Session: https://claude.ai/session/relay-roll-drain-timeout-and-batch-cap |