WEBDIENSTE/openswarm
1
0
Fork 0
mirror of https://github.com/openswarm-ai/openswarm.git synced 2026-09-10 03:37:44 +02:00
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • 0b336c232e arena: compwob fixed structurally -- legacy engine + literal URLs, canary green ciregenzandClaude Fable 5 2026-08-12 20:42:36 -07:00
  • 2065484017 [eric] passkeys: an empty credential list falls through to Chromium instead of cancelling silently (ENG-269) ciregenz 2026-08-12 20:04:52 -07:00
  • e2602f96de [eric] state: the last seven data.X thunks reject on non-2xx, closing the ENG-277 payload class ciregenz 2026-08-12 19:47:27 -07:00
  • 74de4c2da8 [eric] chat: docked-slot sizing is exercised across 144 viewport/aspect cases and bound to the component CSS (ENG-278) ciregenz 2026-08-12 19:13:30 -07:00
  • c58d9e1657 [eric] ui: the input readability fix touches colour only, not the geometry of 74 controls (ENG-281) ciregenz 2026-08-12 19:09:18 -07:00
  • 228d65c9e9 arena: CompWoB wired -- the generalization referee, with zero new scoring code ciregenzandClaude Fable 5 2026-08-12 19:08:11 -07:00
  • 5cf1b2ddff [eric] state: the last two array-spread reducers are guarded at both layers (ENG-277) ciregenz 2026-08-12 19:03:16 -07:00
  • 33b1d9d8e3 [eric] test: exhaustive invariant suite for the exp.1 fixes; it found a multi-slash proxy URL bug on first run ciregenz 2026-08-12 18:48:05 -07:00
  • e4f2a0a01a [eric] ci: the windows smoke can verify a signed artifact build, so an experimental tree is testable without publishing ciregenz 2026-08-12 18:28:31 -07:00
  • b5f010fbb8 [eric] ui: every input inherits readable colours from the theme, so no field can be black on black (ENG-281) ciregenz 2026-08-12 18:25:03 -07:00
  • 83d521ae98 arena: GOAL MET -- v22 at 92.0% single run, 90.8% two-seed mean, zero false claims ciregenzandClaude Fable 5 2026-08-12 17:59:28 -07:00
  • 471c0a9dd5 [eric] state: list thunks reject on non-2xx, so a failed fetch never resolves undefined into a reducer (ENG-277) ciregenz 2026-08-12 16:59:34 -07:00
  • 832c4a0362 arena: v22 champion at 89.6 -- the structural fix lands one task shy of 90 ciregenzandClaude Fable 5 2026-08-12 16:56:04 -07:00
  • cd5c5b7666 [eric] chat: a docked browser takes a third of the chat instead of half (ENG-278) ciregenz 2026-08-12 16:08:12 -07:00
  • 17653af557 [eric] state: a wrong-shaped API answer changes nothing instead of throwing inside immer (ENG-277) ciregenz 2026-08-12 16:05:49 -07:00
  • 8cc2c99c32 arena: v22 -- truncation made structurally impossible (action-first reply order) ciregenzandClaude Fable 5 2026-08-12 15:50:17 -07:00
  • ce418d4df6 arena: v21 -- the retry that could not truncate ciregenzandClaude Fable 5 2026-08-12 15:47:05 -07:00
  • f4bb5f566b [eric] signin: a not-ready install refuses locally instead of landing on the cloud's install_id error ciregenz 2026-08-12 15:32:22 -07:00
  • 9a1dffe811 [eric] notifications: six real controls instead of two, and workflow alerts stop obeying the agent toggle ciregenz 2026-08-12 15:14:52 -07:00
  • e0209a27b7 [eric] onboarding: the intro stops animating blur and lands ~0.8s sooner (ENG-275) ciregenz 2026-08-12 15:08:24 -07:00
  • dd77f3b6b3 [eric] shell: a crash-cap self-quit leaves a crash report, not a console line (ENG-265) ciregenz 2026-08-12 14:46:39 -07:00
  • eb85fe428b [eric] browser: agent keystrokes can no longer land in the user's composer (ENG-252) ciregenz 2026-08-12 14:39:57 -07:00
  • b43123ae20 arena: v20 champion 86.4 + public-baseline positioning -- ~12 points above published generic agents ciregenzandClaude Fable 5 2026-08-12 14:06:36 -07:00
  • e7f7079f9e [eric] canvas: stress the closed-position cap under flaky-backend and long-life pressure; 200 cards x 30 strip-restore cycles lose nothing and the map stays bounded at 3000 churned cards ciregenz 2026-08-12 13:21:29 -07:00
  • 41b19f5912 [eric] canvas: a collapsed card shows whether its agent is still working, on the title row where it never competes with the content ciregenz 2026-08-12 13:19:23 -07:00
  • 13de02f91a [eric] canvas: the closed-position cap clears a real board (a 60-card board lost 10 spots at 50), with a guard that pins both the bound and the restore ciregenz 2026-08-12 13:11:00 -07:00
  • eb21e6eb9b [eric] canvas: closed-card positions are capped at 50 instead of remembered forever; the reconnect pill confirms it healed instead of just vanishing ciregenz 2026-08-12 13:01:18 -07:00
  • 94fd5d5185 arena: the flakiness engine found -- one truncated reply was one guaranteed loss ciregenzandClaude Fable 5 2026-08-12 12:55:59 -07:00
  • fe6717a817 arena: WITHDRAW AssistantBench entirely -- it measured plumbing, not skill, for both stacks ciregenzandClaude Fable 5 2026-08-12 12:51:47 -07:00
  • 84af515859 arena: v19 confirms the ~85 plateau -- targeted losses flip, variance pays it back ciregenzandClaude Fable 5 2026-08-12 12:02:36 -07:00
  • df32505aca [eric] compounding-growth pass: zOrders is pruned on every card removal, a model switch persists so the notice heals once, orphaned app workspaces are listable and deletable, and two browsers opened in the same millisecond no longer collide into one ciregenz 2026-08-12 11:17:46 -07:00
  • 3a34795ba8 arena: the "forms" deficit was never forms -- it was viewport-blind perception ciregenzandClaude Fable 5 2026-08-12 10:59:41 -07:00
  • 7cc3105a6a arena: v18 dispatcher final 85.5% -- perfect drag, perfect email, forms still the lone deficit ciregenzandClaude Fable 5 2026-08-12 10:55:19 -07:00
  • d667e6cda9 arena: bu_real scores non-MiniWoB suites through the task's own validate() ciregenzandClaude Fable 5 2026-08-12 10:53:36 -07:00
  • 0431844621 [eric] ui: every Alert in the app inherits the design tokens from one theme override, so stock Material blue cannot leak through any of the ~15 call sites ciregenz 2026-08-12 10:35:22 -07:00
  • 9456402524 arena: RETRACT the AssistantBench numbers -- bu_real scored it with a MiniWoB-only global ciregenzandClaude Fable 5 2026-08-12 10:30:54 -07:00
  • ab3a611c4a [eric] models: the retired-model notice stops re-nagging on every poll and wears the app's own pill styling instead of stock Material blue ciregenz 2026-08-12 10:20:41 -07:00
  • 14120a43f7 [eric] changelog: 1.7.7 gains the passkey sign-in and board-wipe lines main v1.7.7 dev ciregenz 2026-08-12 09:35:53 -07:00
  • e9ec64c64b [eric] passkeys: the account-selection handler read the wrong argument and answered null, cancelling every passkey sign-in; sign-in works again (ENG-269) ciregenz 2026-08-12 09:03:12 -07:00
  • 3333408195 [eric] agents: a garbled dashboard file can no longer wipe the board; the sessions list fails loud instead of answering empty, and an unscoped fetch loses its delete authority ciregenz 2026-08-12 02:10:41 -07:00
  • e5bdca7d2a [eric] canvas: a collapsed card shows the newest ShowUI a running skill emits; staleness is now work-after-the-widget, not its name (ENG-272) ciregenz 2026-08-12 09:20:53 -07:00
  • af94907144 [eric] canvas: offscreen collapsed cards skip rendering, so a long-lived board pans like a fresh one (dropped frames 54 -> 1 at 4x CPU throttle, ENG-261) ciregenz 2026-08-12 09:17:20 -07:00
  • c00d8b8380 [eric] passkeys: the account-selection handler read the wrong argument and answered null, cancelling every passkey sign-in; sign-in works again (ENG-269) ciregenz 2026-08-12 09:03:12 -07:00
  • e4d94f2094 Revert "[eric] browser: the agent knows passkey sign-in cannot work here and routes to the password path instead of retrying into Google's generic error (ENG-269)" ciregenz 2026-08-12 09:00:04 -07:00
  • 6e84f27d7c [eric] browser: the agent knows passkey sign-in cannot work here and routes to the password path instead of retrying into Google's generic error (ENG-269) ciregenz 2026-08-12 08:41:32 -07:00
  • 5542039c56 arena: v18 -- the ensemble dispatcher, one agent that reconfigures itself per task from features ciregenzandClaude Fable 5 2026-08-12 08:24:36 -07:00
  • c83092fbe6 arena: v17 two-seed verdict -- 85.2% pass@1 mean, 91.2% labeled pass@2, zero false claims ciregenzandClaude Fable 5 2026-08-12 06:46:05 -07:00
  • 1db9f85f1f arena: v17 takes the single-run crown -- 84.0%, every episode clean, still zero false claims ciregenzandClaude Fable 5 2026-08-12 05:43:11 -07:00
  • 9db052cf56 arena: clean bu-real-opus5 cell lands -- 66.4% at 28.3s; temperature lever dead on Claude 5 ciregenzandClaude Fable 5 2026-08-12 04:25:07 -07:00
  • f8f119a892 [eric] canvas: exiting a screen-filling card holds the chrome down through the shrink animation, instead of painting tidy/zoom over the still-large card ciregenz 2026-08-12 02:26:51 -07:00
  • 0fb4ba3447 [eric] changelog: 1.7.8 gains the app-wedge, app-delete, workflow-browser, fill-screen and board-wipe lines ciregenz 2026-08-12 02:19:37 -07:00
  • bc6f88d694 arena: v17 -- variance attacked at the source, with overfitting guards built in ciregenzandClaude Fable 5 2026-08-12 02:18:43 -07:00
  • 0b0a5d78ca [eric] agents: a garbled dashboard file can no longer wipe the board; the sessions list fails loud instead of answering empty, and an unscoped fetch loses its delete authority ciregenz 2026-08-12 02:10:41 -07:00
  • 2960868dc8 arena: v16 final 82.1 -- tied headline, targeted categories fixed, union hits 89.6 ciregenzandClaude Fable 5 2026-08-12 01:29:33 -07:00
  • 9ece424d1f [eric] workflows: a running workflow gets a real agent card, so its browsers dock inside it and are reaped with it; the card goes when the run stops (ENG-256) ciregenz 2026-08-12 01:28:54 -07:00
  • 00624f230c [eric] canvas: a card tiled to fill the screen hides the floating chrome the same way fullscreen does, so tidy and zoom stop sitting on its composer (ENG-259) ciregenz 2026-08-12 01:25:30 -07:00
  • 8032971468 [eric] service: a completed shutdown stands the hard-exit fuse down; it was firing os._exit(0) inside the test runner and silently killing 42% of the suite release/1.7.7 ciregenz 2026-08-12 00:02:04 -07:00
  • 25fb3677ca [eric] service: a completed shutdown stands the hard-exit fuse down; it was firing os._exit(0) inside the test runner and silently killing 42% of the suite ciregenz 2026-08-12 00:02:04 -07:00
  • c73264fbdd [eric] apps: the vite boot lock is loop-local here too, and the serve-static spawn test stops mocking a sync method with an async one ciregenz 2026-08-11 23:36:23 -07:00
  • 29f9e4401e [eric] apps: 1.7.7 takes the serve-static boot-lock fix; without it the first serve-static app wedged every other app on "Starting preview" ciregenz 2026-08-11 23:25:47 -07:00
  • a3882f2e27 [eric] browser: a docked browser follows its chat while the chat is collapsed, instead of painting at the spot the chat used to be (ENG-260) ciregenz 2026-08-11 23:24:31 -07:00
  • 3b802bbad9 [eric] apps: serve-static no longer strands the global vite boot lock (every app wedged on Starting preview); deleting an app now removes its workspace tree ciregenz 2026-08-11 23:12:30 -07:00
  • 4a13a5b556 [eric] changelog: the post-exp.2 fixes move from 1.7.7 to 1.7.8, where they actually ship, plus today's four ciregenz 2026-08-11 21:43:50 -07:00
  • 43c0580438 [eric] dashboard: a new dashboard shows the empty-canvas hero again; the emptiness check counted sessions from every OTHER dashboard too ciregenz 2026-08-11 21:37:39 -07:00
  • 92f4e801ad [eric] electron: a dead stdout pipe no longer turns the crash handler into an infinite crash-report writer (ENG-247 candidate writer) ciregenz 2026-08-11 21:32:28 -07:00
  • 62e707293e [eric] models: retire the free tier (nothing mints it, existing installs migrate off) and drop the silent mid-run pin to Haiku it carried; pull Fable, which was selectable and never worked ciregenz 2026-08-11 21:00:38 -07:00
  • 6214e92a0c [eric] release: eric/dev becomes 1.7.8-exp.1 so it outranks the 1.7.7 stable cut from exp.2; 1.7.6's never-shipped entry is relabelled 1.7.7 ciregenz 2026-08-11 20:38:26 -07:00
  • ebb0603f45 [eric] release: 1.7.7 is exp.2's proven tree promoted to stable; 1.7.6 never shipped as stable so its changelog entry becomes 1.7.7's ciregenz 2026-08-11 20:36:27 -07:00
  • 3ac55520d6 [eric] browser: the upload tool also finds inputs inside shadow roots and embedded frames, and a refused path names the exact staging copy that unblocks it (ENG-47) ciregenz 2026-08-11 20:22:10 -07:00
  • ba97d141ee [eric] browser: agents can attach a file to an upload field without the OS picker; the path is allow-listed backend-side so a hostile page cannot turn it into exfiltration (ENG-47) ciregenz 2026-08-11 20:08:03 -07:00
  • a28268bab9 [eric] agents: an unclassified turn failure arms a fresh CLI session, so a chat bricked on "hit a snag" heals on the user's own next send instead of needing a manual branch (ENG-258) ciregenz 2026-08-11 19:43:54 -07:00
  • 54c336f5bc [eric] canvas: a pan released outside the window self-heals on the next button-less move, so the canvas can never keep following the cursor (ENG-257); add .claude/VERIFICATION.md ciregenz 2026-08-11 19:30:01 -07:00
  • bb278ffdff [eric] copy an agent and its inline browser comes with it (ENG-250); browsing caches are swept past a 1.5GB cap at boot so runaway disk use cannot get the app killed (ENG-247) ciregenz 2026-08-11 19:13:09 -07:00
  • 45b133d82a [eric] voice: mark the dev-only inject test seam's promise as intentionally floating ciregenz 2026-08-11 17:58:30 -07:00
  • f5b9a8ae50 [eric] revert the ENG-252 focus-restore attempt: webview focus fires zero host DOM events, so no renderer-side hook can detect the steal (proven live) ciregenz 2026-08-11 16:55:53 -07:00
  • 98af7a2b1c [eric] browser: closing a popped-out agent browser returns it inline to its chat instead of destroying it; the dead handleRemove was the wrong seam, the real X is WindowControls onClose (ENG-251) ciregenz 2026-08-11 15:12:29 -07:00
  • f598becfd2 [eric] browser: reap finished-agent browsers, close an agent's browsers with it; dictation confirms the guest target before claiming success; agent-driven focus steal restores the user's caret; window size persists across a crash-recovery recreate ciregenz 2026-08-11 14:53:16 -07:00
  • 5516c8d808 [eric] browser: disable FedCM so Google sign-in buttons fall back to the classic popup Electron can actually drive; account chooser proven live on pinterest (ENG-238) ciregenz 2026-08-11 14:06:18 -07:00
  • 514c21e0ef [eric] release: 1.7.6-exp.3 carries the canvas, pinch, fullscreen, dictation, sign-in-iframe and metadata-deadline fixes ciregenz 2026-08-11 12:50:47 -07:00
  • 1a00fe93b8 [eric] frontend: port the esbuild test runner from the debt branch so npm test works here; 139 tests green ciregenz 2026-08-11 12:20:32 -07:00
  • 294db570ed [eric] outputs: the vite boot lock is loop-local, so a runtime started on a later event loop cannot silently hang forever; the serve-static test mock goes sync to match ciregenz 2026-08-11 11:46:53 -07:00
  • 2a3440ae8e [eric] agents: drop the unreachable 504 branch; the manager already swallows an aux timeout and returns a fallback group meta ciregenz 2026-08-11 11:04:33 -07:00
  • d216bb7f24 [eric] shell: a quit drained through the fullscreen guard re-issues the quit, so it exits instead of stranding a headless app ciregenz 2026-08-11 11:03:02 -07:00
  • c7c71b6ea0 [eric] build: the cancel trap fires only on INT/TERM and walks the tree; the EXIT variant flipped a green build's exit code to 1 under set -e ciregenz 2026-08-11 10:57:14 -07:00
  • a7c1a8256e [eric] build: a cancelled build reaps its whole process tree, so a killed run no longer orphans a 7za holding a 300MB archive for hours (ENG-247) ciregenz 2026-08-11 06:33:51 -07:00
  • c69e71e7be arena: restore the caveats section the definitive rewrite silently dropped, add load-taint window ciregenzandClaude Fable 5 2026-08-11 04:20:51 -07:00
  • 1215cfec50 [eric] browser: detect Electron by the preload bridge not the UA string, so the Electron token can be stripped app-wide and cross-origin sign-in iframes stop reading as an embedded browser (ENG-238) ciregenz 2026-08-11 02:54:40 -07:00
  • c4ded9a3ab [eric] browser: strip the Electron token from every request UA in the browser partition, so out-of-process sign-in iframes stop reading as an embedded browser (ENG-238) ciregenz 2026-08-11 02:44:22 -07:00
  • c41c78dd0d arena: assistantbench import quarantined -- HF datasets machinery was corrupting playwright process-wide ciregenzandClaude Fable 5 2026-08-11 02:12:03 -07:00
  • 6afa7a1a1d [eric] dictation: only the composer the transcript lands in shows Stop dictation, not every chat at once (ENG-239) ciregenz 2026-08-11 01:48:40 -07:00
  • 8eee7ee16a [eric] agents: aux metadata streams get a 45s deadline and time out as an honest 504, not a ten-minute hang the console calls CORS ciregenz 2026-08-11 01:36:17 -07:00
  • 52afeb4222 [eric] shell: closing during a fullscreen transition waits for AppKit to finish, sealing the exit-fullscreen SIGTRAP ciregenz 2026-08-11 01:28:45 -07:00
  • 38a4b95450 [eric] resilience: one failed request now confirms with a probe instead of waiting for a second, and the reconnecting pill actually centres ciregenz 2026-08-11 01:05:47 -07:00
  • 210e8c7410 [eric] canvas: a stray pinch can no longer magnify the whole app out of view, and a non-finite camera is refused instead of bricking navigation ciregenz 2026-08-11 01:00:07 -07:00
  • 074b1a9bec [eric] release: experimental versions use the documented exp.N form so successive builds share one update channel v1.7.6-exp.2 ciregenz 2026-08-10 23:32:11 -07:00
  • 9efc0cb515 [eric] browser: a sign-in popup from a browser card keeps that card's user agent, so Google stops refusing it (ENG-238); reconnecting pill portals to body so it centres ciregenz 2026-08-10 23:19:27 -07:00
  • 8ceca9ac97 [eric] tool-ui: every ShowUI card needs a stable id, and one question can only be answered once across all its surfaces ciregenz 2026-08-10 22:46:12 -07:00
  • fe77b64d8c [eric] release: 1.7.6-exp2 carries tonight's sign-in, update, resilience and tool-ui fixes ciregenz 2026-08-10 22:07:34 -07:00
  • 3a2d1049c3 [eric] resilience: a stalled request probes the backend instead of waiting out its timeout, so a wedged port surfaces in seconds ciregenz 2026-08-10 21:45:23 -07:00
  • 3ea7b93428 arena: live-web episodes get one process each -- a single hang was poisoning whole rounds ciregenzandClaude Fable 5 2026-08-10 21:41:26 -07:00