Eric
ca5d69301d
[eric] app builder: fix windows-built apps failing by always creating the workspace .env (the windows build dropped the .env.example it seeds from) and writing app files as utf-8 (windows cp1252 choked on arrows and chess pieces); stays on 1.1.69 to rebuild only the windows installer
2026-05-26 17:40:51 -07:00
Eric
8c5b3bdf64
[eric] ui: fix the windows guide cursor jumping instead of gliding (it was being rebuilt from scratch every frame) and remove the new chat / history pills above the composer; bump 1.1.69
2026-05-26 15:41:13 -07:00
Eric
440a3e22f6
[eric] windows: fix "backend crashed" by making the app pick a port that is genuinely free so it stops colliding with another program already on port 8324; bump 1.1.68
2026-05-26 14:17:12 -07:00
Eric and Claude Opus 4.7
45a97aa05f
[eric] windows: stop nsis upgrades hitting "cannot be closed" by reaping app builder children on quit and fixing the dead orphan-killer; bump 1.1.67
...
- root cause: window-all-closed called killBackend() (windows: taskkill /F) BEFORE before-quit could POST /shutdown-all, so the backend died before running stop_all(), orphaning the bundled vite node.exe; a running node.exe locks its own image at resources\node\x64\node.exe, so the next nsis upgrade cannot overwrite it and surfaces appCannotBeClosed via the file-lock retry path (extractAppPackage.nsh / installUtil.nsh), not the name-based app-running check
- windows-only because mac's killBackend sends SIGTERM, letting uvicorn run its lifespan shutdown -> stop_all(); only since 1.1.64 because app builder first spawned these children on windows in bf6d7a9noreply@anthropic.com >
2026-05-26 13:30:05 -07:00
Eric and Claude Opus 4.7
c0d0e012cc
[eric] outputs/quit: reap app builder child processes before killing the backend on windows so nsis upgrades stop hitting "cannot be closed"; bump 1.1.67
...
- root cause: window-all-closed called killBackend() (windows: taskkill /F) BEFORE before-quit could POST /shutdown-all, so the backend was dead before it could run stop_all(), orphaning the bundled vite node.exe; a running node.exe locks its own image at resources\node\x64\node.exe, so the next nsis upgrade cannot overwrite it and surfaces appCannotBeClosed via the file-lock retry path (extractAppPackage.nsh / installUtil.nsh), not the name-based app-running check
- windows-only because mac's killBackend sends SIGTERM, which lets uvicorn run its lifespan shutdown -> stop_all(); only since 1.1.64 because app builder first started spawning these children on windows in bf6d7a9noreply@anthropic.com >
2026-05-26 13:19:11 -07:00
Eric and Claude Opus 4.7
cf74ada3c2
[eric] auth: generate installation_id at backend startup so first-launch sign-in always has a non-empty install_id; bump 1.1.66
...
- installation_id was created lazily on the first analytics submission; 1.1.64 removed the workflows startup poller, which was the early backend ping that used to generate it, so on a clean install the sign-in window built its google/email oauth url with an empty install_id and the cloud rejected the start
- now generated in main.py at the same pre-bind moment as the auth token, so GET /api/settings carries it from the very first fetch (no dependency on analytics timing)
- idempotent uuid4 hex, only written when missing; lazy path kept as fallback; platform-agnostic so mac is unchanged except the id exists slightly earlier
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com >
2026-05-26 12:25:31 -07:00
Eric
40b7c19b75
[eric] onboarding cursor: restore writePos-after-spring order so mac stays byte-identical (windows still eases via no-op controls + css transition); bump 1.1.65
2026-05-26 11:17:01 -07:00
Eric
3509c08057
[eric] workflows: remove scheduled-tasks end-to-end (frontend pages+slice, backend app+routes, electron poller), keep dock chat; bump 1.1.64
2026-05-26 11:12:29 -07:00
Eric
bf6d7a9235
[eric] app-builder: cross-platform workspace runtime so apps build and run on windows (resolve bash/python/npm, venv Scripts layout, bundled-node PATH, online npm retry)
2026-05-26 04:18:38 -07:00
Eric
9605454ebf
[eric] windows: onboarding cursor eases via cursorStore-driven css transition (snaps while tracking, eases on moveTo) to match the mac spring
2026-05-26 04:16:57 -07:00
Eric
529d350cae
[eric] windows 1.1.63: drop the textarea + file-input squirrel-era ablations, restore contentEditable @-mention UI and the attach
...
paperclip on windows, plus a cubic-bezier cursor transition
2026-05-26 02:42:56 -07:00
Eric
da61377d8c
[eric] windows 1.1.62: cursor and popups follow target on windows via store-driven style.transform
2026-05-26 02:10:56 -07:00
Eric
31a90502c7
[eric] windows 1.1.61: restore AgenticCursor render on Windows now that the framer-motion shim makes motion.div safe
2026-05-26 01:42:05 -07:00
Eric
fc8c35e5e0
[eric] windows 1.1.60: revert installer to nsis to test if squirrel was the renderer-crash trigger
2026-05-26 01:20:20 -07:00
Eric
bb2de72f6f
[eric] windows 1.1.59: production webpack and framer-motion shim
2026-05-26 00:29:17 -07:00
Eric
7407e6cb37
[eric] windows 1.1.58: three fixes — (a) revert file-input restore: <input type="file"> is ALSO a Windows commit-phase crasher just like the contentEditable was, both must stay ablated, drag-and-drop still works; (b) nuclear-disable OnboardingRoot on Windows since the AgenticCursor visual ablation alone was insufficient (Onboarding panel/director/popups have additional Framer Motion + portal subtrees that segfault on commit) — user loses the guided tour on Windows but the rest of the app works; (c) remove iframe sandbox from BrowserCard so sites that need full-document access can render (Google specifically still appears grey because they run JavaScript anti-iframe detection that blanks document.body when window.top !== window.self, which header stripping cannot defeat; most non-Google sites should render fine now); Mac path untouched
2026-05-25 23:47:14 -07:00
Eric
aff9ccf98a
[eric] windows 1.1.57: four fixes — (a) textarea Enter-to-send now actually clears the input (missed editor.innerHTML='' at ChatInput.tsx:181 in the post-send path); (b) textarea draft restore (useDraftLoad) routes through .value instead of innerHTML on Windows so a saved draft repopulates the input on remount; (c) restore the <input type="file"> + attach button on Windows since the original 1.1.54 crash root cause was the contentEditable TSF init not the file input — replacing contentEditable with textarea was the actual fix, file input was over-ablated; (d) relax CSP frame-src to allow http:/https: so BrowserCard iframe can render external sites; (e) gate AgenticCursor's portaled Framer Motion infinite-loop animation on Windows so onboarding panel commit no longer 0xC0000005 segfaults; Mac path untouched
2026-05-25 23:22:06 -07:00
Eric
3903766e96
[eric] windows 1.1.57: (a) relax CSP frame-src to allow http: and https: so BrowserCard iframe can render external sites (our own app-level CSP was the second-layer blocker after the upstream XFO/CSP frame-ancestors we already strip in main.js — together they let the iframe load Google et al); (b) gate AgenticCursor's portaled Framer Motion infinite-loop scale + drop-shadow node on Windows so the onboarding panel commit phase no longer segfaults from that subtree (visual cursor disabled on Windows, imperative handle still works for AC runtime); Mac path untouched
2026-05-25 23:16:14 -07:00
Eric
0076909145
[eric] windows 1.1.56 hotfix: webRequest.onHeadersReceived filter uses 'subFrame' (Electron camelCase) not 'sub_frame' (Chrome extension snake_case) — wrong name threw "Invalid type sub_frame" synchronously during app.whenReady, became an unhandledRejection, prevented the app from booting on 1.1.55
2026-05-25 22:51:34 -07:00
Eric
b73f508755
[eric] windows 1.1.55: (a) textarea-aware editor read/write/clear helpers so Enter-to-send works on Windows where EditorSurface uses textarea instead of contentEditable (handleSend was reading .textContent which is empty on textarea); (b) gate BrowserCard webview behind !Windows so dashboard hydration no longer segfaults from webview tag commit (iframe fallback is now usable for most sites since we also strip X-Frame-Options and CSP frame-ancestors on sub_frame loads via session.webRequest.onHeadersReceived, scoped to http/https types:sub_frame so OAuth popups and deep links keep their security headers); (c) gate onboarding video autoPlay on Windows so panel mount does not trigger the autoplay+animation crash; Mac path untouched everywhere
2026-05-25 22:29:17 -07:00
Eric
518d0574dd
[eric] windows 1.1.54 ablate (fix): correctly platform-gate the file-input/attach-button ablation (1.1.53 mistakenly disabled them on both Mac and Windows because the file was already wrapped in {false && ...} from a prior diagnostic — Mac now renders the full attach UI again, Windows still skips it); textarea-instead-of-contentEditable already correctly platform-gated since 1.1.53
2026-05-25 21:59:19 -07:00
Eric
d85d9c2c10
[eric] windows 1.1.52 ablate: swap contentEditable div for textarea + hide <input type="file"> on Windows so the commit-phase 0xC0000005 (Chromium 144 + Electron 40 native crashers) cannot mount; Mac path untouched so @-mention rich-UI and full attach button stay working there; UA-gated at module load via navigator.userAgent.includes('Windows') so zero runtime branch cost; if Windows chat opens without crash one of these was the trigger and we narrow in v1.1.53 by re-enabling each
2026-05-25 21:53:04 -07:00
Eric
e2ea5e03cf
[eric] windows 1.1.51 diag: render logs on every ChatInputToolbar child (ModeControl, ModelPickerMenu, ThinkingLevelControl, ToolbarActions) so the next 0xC0000005 crash narrows from "ChatInputToolbar last" to the exact toolbar child that commits before the segfault
2026-05-25 21:27:14 -07:00
Eric
06e7c54f57
[eric] windows 1.1.50 diag: render logs on every ChatInputView child (ChatInputToolbar, AttachmentChips, ChatInputOverlays, SendBlockBanner) so the next 0xC0000005 crash trace shows which sibling commits after EditorSurface — spellCheck=false on the contentEditable did not stop the crash so the trigger is either later in ChatInputView's child render order or in the JSX after EditorSurface's render log; bump for Squirrel fresh-fetch
2026-05-25 21:04:46 -07:00
Eric
664faa1033
[eric] windows 1.1.49: spellCheck={false} on contentEditable div + ChatInputView and EditorSurface render logs + version bump so Squirrel actually fetches a fresh signed binary (force-retagging v1.1.44 was reusing the cached install on every iteration); tag v1.1.49
2026-05-25 20:04:37 -07:00
Eric
83a042d662
[eric] windows: drop invalid autoCorrect + autoCapitalize attrs from contentEditable div in EditorSurface — diag logs pinpointed ChatInput as the last child rendered before every 0xC0000005 segfault, and these attributes are spec-defined for input/textarea only so React forwarded them onto a div where Chromium 144 + Windows IME / spellchecker engages a native code path that segfaults during commit; spellCheck stays for spelling underlines; retag v1.1.44
2026-05-25 19:47:55 -07:00
Eric
294d1a91ed
[eric] diag: add targeted logs for each top crash-suspect (boot env snapshot for React-prod-mode + tree-shaking checks, wrapped lazy() loader with requested/loaded/failed logs for chunk-split race, [diag][ErrorBoundary] prefix on componentDidCatch so in-tree throws surface in stderr alongside window.onerror); retag v1.1.44
2026-05-25 19:26:40 -07:00
Eric
1ec3d473f3
[eric] diag: webpack --mode=development for packaged frontend + render logs on every AgentChat child (ContextDrawer / MessageBubble / StreamingBubble / CompactionMarker / ToolCallBubble / ToolGroupBubble / ApprovalBar / ChatInput / MessageActionBar / ErrorSlime) so the next 0xC0000005 crash trace shows the exact last component that committed before the segfault since http origin already disproved file://; retag v1.1.44
2026-05-25 19:20:16 -07:00
Eric
0c2cf3fc20
[eric] diag: build packaged frontend with webpack --mode=development since http://127.0.0.1 origin did not stop the 0xC0000005 crash (file:// theory disproved), and the only remaining variable between crash-free dev mode and crash-on-every-chat packaged mode is the production-vs-development webpack bundle (minification, terser, React prod mode, removed dev safety checks); larger bundle for one release to confirm minification/prod-bundle is the trigger, will revert and engineer a surgical webpack config once isolated; retag v1.1.44
2026-05-25 19:16:44 -07:00
Eric
0bec1ea268
[eric] windows: serve packaged frontend via embedded http://127.0.0.1 server instead of file:// since dev mode (http origin) never crashed while packaged (file://) consistently 0xC0000005 segfaulted on chat/dashboard mount, confirming file:// origin is the actual native-crash trigger on Electron 40 CastLabs Chromium 144; revert disableHardwareAcceleration to keep GPU perf since the fault was never GPU-side; retag v1.1.44
2026-05-25 18:40:06 -07:00
Eric
9a6956477f
[eric] windows: app.disableHardwareAcceleration() since dropping the gpu-blocklist override alone did not stop 0xC0000005 renderer crashes on chat/dashboard mount, so the native segfault is in the GPU process itself not just blocklist-bypass paths; software rasterization is slower but neutralizes the entire GPU-side native-crash class; win32 only, mac is unaffected; retag v1.1.44
2026-05-25 18:15:03 -07:00
Eric
bfe59e52b0
[eric] windows: drop ignore-gpu-blocklist + enable-gpu-rasterization + enable-zero-copy since Crashpad confirms every renderer crash is STATUS_ACCESS_VIOLATION 0xC0000005 with no JS cause, classic blocklisted-driver segfault pattern on Electron 40 CastLabs Chromium 144; respecting the blocklist lets Chromium fall back to software rasterization on affected GPUs; retag v1.1.44
2026-05-25 17:49:12 -07:00
Eric
099a0c69e7
[eric] diag: max-coverage logs (crashReporter for native minidumps, main-process uncaughtException + unhandledRejection + child-process-gone + preload-error, full render-process-gone details, safe sendToRenderer, every ipcMain.handle entry, BrowserCard / WorkflowCard render, AgentChat before-jsx) so the next renderer crash leaves a complete trace; retag v1.1.44
2026-05-25 17:29:40 -07:00
Eric
aa644d2d2a
[eric] fix: WebSocketManager.connect diag log read this.options.sessionId on a class that stores the field as this.sessionId directly, so every WS connect threw TypeError and triggered the Windows native renderer crash on every chat / dashboard hydration; retag as v1.1.44 since the chat-spawn root cause was already closed by the v1.1.44 preload race fix and my v1.1.45 instrumentation regressed it
2026-05-25 16:49:01 -07:00
Eric
be3808fa5c
[eric] diag: add global window.onerror + unhandledrejection handlers in Main.tsx so packaged-build stderr captures the stack trace of the "Cannot read properties of undefined (reading 'sessionId')" throw firing inside AgentChat's render path (bundle.js:2 alone is uninformative without source maps)
2026-05-25 16:27:39 -07:00
Eric
d27f289cea
[eric] windows: fix UA-spoof leak onto recreated main window + add diag logs across preload, createWindow, recreateMainWindow, resumeSession, launchAndSendFirstMessage, AgentChat, WebSocketManager so packaged-build stderr captures the chat-spawn crash trail
2026-05-25 16:06:58 -07:00
Eric
e7a09c9568
[eric] windows: fix preload race that crashed renderer on chat spawn
...
preload.js awaited two ipcRenderer.invoke calls before exposing
window.openswarm. Renderer code running in the gap saw the global
undefined: BrowserCard fell back to iframe mode (grey panel on sites
with X-Frame-Options), and chat-spawn calls to getAuthToken threw at
mount, crashing the renderer on Electron 40 + Windows.
Only bit Windows post-lazy-boot (ddd82ef
2026-05-25 15:45:37 -07:00
Eric
6777d45d1c
windows: stop killing app on renderer crash; recreate window instead of reload
...
The render-process-gone handler called mainWindow.reload() to recover from
renderer crashes (e.g. WebGL / native / OOM in chat-spawn flows). On the
Electron 40 CastLabs build the reload re-registers a session/webview observer
that Chromium base/observer_list.h:318 refuses to add twice. The DCHECK
aborts the entire main process with exit 3, taking the app down whenever the
renderer dies.
Recreate the BrowserWindow instead: a fresh webContents has no duplicate
observers to add. The closed handler is identity-checked so the old window
teardown does not null out the new mainWindow reference.
When the 3-in-60s crash cap trips we now surface a native dialog (Reload /
Quit) instead of leaving a stuck dead window. Native dialog runs outside the
render-process-gone call stack, so its Reload button cannot itself reproduce
the DCHECK.
Bumps to 1.1.43 (re-cutting the deleted release draft).
2026-05-25 15:14:29 -07:00
ciregenz
bcabcca407
[eric] workflows: scheduling is a chat view (bounded height, composer docked) so it stops collapsing tiny
2026-05-25 12:51:47 -07:00
ciregenz
c5d2fe512f
[eric] workflows+dashboard: card width matches chat, Cmd/Ctrl+A select-all, compact scheduling + consistent cancel/title
2026-05-25 12:45:59 -07:00
ciregenz
846ee3bb01
[eric] workflows: auto-fit card height, drop stray step bands, fix active-poll 401 + view dead-ends
2026-05-25 12:22:25 -07:00
ciregenz
85646bc556
[eric] workflows: ignore saves card, scheduling chat typeable + default model, accent color, drop step box
2026-05-25 11:50:12 -07:00
ciregenz
a137eba7c0
[eric] workflows: preview header divider, HITL-gold schedule button, model/mode/secs subtitle, slicker calendar
2026-05-25 11:16:30 -07:00
ciregenz
db4b944b0c
[eric] workflows: compact expandable steps in preview so schedule prompt + buttons stay visible
2026-05-25 10:32:12 -07:00
ciregenz
083d70def5
[eric] workflows: merge scheduled-tasks onto dev, re-home dashboard + fix backend imports
2026-05-24 23:41:53 -07:00
ciregenz
8ee037d770
[eric] release: bump version to 1.1.42
2026-05-24 16:35:38 -07:00
ciregenz
6cf2606666
[eric] merge: fold eric/cleanup abstraction refactor into dev (resolve 9, re-home edits)
2026-05-24 16:24:59 -07:00
ciregenz
5fb4d02648
[eric] ui: docked tour collapses to a minimal right-edge tab; restore serif font
2026-05-24 12:33:22 -07:00
ciregenz
edec6e0f9c
[eric] ui: onboarding skip-to-sidebar + global count; sans fallback; memoize hot selectors
2026-05-24 11:03:30 -07:00
ciregenz
6e2d4e94aa
[eric] actions: scope MCP service rules per integration; fixes mislabeled @ commands
2026-05-24 02:40:35 -07:00
Eric