* feat(server): add Gitea, Forgejo, and Codeberg support for change request watches
Implement the Gitea-family ChangeRequestSource using cwd-free tea login list and tea api --login -i requests, inspecting HTTP status codes directly on stdout/stderr rather than treating exit code 0 alone as success.
Reuse the shared per-host bounded poller for concurrency limits, 429 Retry-After pauses, and exponential backoff. Combine commit statuses and Actions tasks (including approval-held runs via need_approval and cross-repo waiting shadows) with mid-pagination headSha validation, and paginate comments, reviews, and inline review comments with bounded fan-out and quiet-poll caching.
* fix(server): address review findings on Gitea watch pagination, check sharing, and runs fetch
- Read /pulls/:index/reviews/:id/comments in a single request instead of a page loop because Gitea ignores page/limit on that endpoint and omits X-Total-Count/Link headers.
- Paginate /commits/:sha/status using the X-Total-Count response header instead of body total_count (which Gitea sets to the page length).
- Extract shared pure Gitea check schemas, shadow detection, rerun selection, and tea login matching into gitea-checks.ts imported by both gitea-service.ts and gitea-change-requests.ts.
- Remove the redundant /actions/runs fetch from the watch adapter because watch checks carry no action_required trait and approval-held tasks/pending shadows already map to pending.
* fix(server): preserve Gitea watch identity, checks, and backoff
* fix(server): route Gitea watches by base URL and retain complete updates
* fix(server): reconcile Gitea checks by job and refresh watch logins
* fix(server): validate Gitea and Forgejo Actions job responses
* fix(server): classify canceled native Actions statuses as neutral
* fix(server): drain failed Gitea polls before releasing host slots
* fix(server): let admitted Gitea polls finish with low quota
* fix(server): normalize Gitea watch URLs and refresh inline replies
* test(server): cover late Gitea rate limits across recovery windows
* feat(server): add multi-forge routing and GitLab support for change request watches
Route change request watch URLs host-first through ForgeResolver (cloud host match plus cached async CLI auth probe) before delegating path canonicalization and polling to the matched forge adapter, while sharing adapter instances with workspace reads.
Persist the resolved forge id on saved watch records (defaulting legacy records to github) so daemon restarts recover active watches directly by forge without probing hosts during outages, and keep unknown-forge records intact rather than aborting boot.
Add the GitLab ChangeRequestSource using cwd-free glab api --hostname calls, bounded per-host polling and rate-limit/error backoff, head-pipeline job/bridge checks with mid-pagination headSha verification, discussions/comments, and observable GitLab approval transition updates without manufacturing a GitHub reviewDecision.
* fix(server): address review findings on GitLab watch rate limits, approvals, and legacy recovery
- Decide rate limits from HTTP status codes only on parsed glab responses so MR titles, comments, or IIDs mentioning 429 or rate limit never trip false host pauses.
- Remove the duplicate in-memory approval baseline and counter in favor of immutable GitLab approval system notes enriched with /approvals counts only on new notes, keeping seenActivity as the single persisted owner.
- Infer forge from URL shape when a persisted record omits forge so GitLab watches survive an older-daemon downgrade/upgrade cycle without probing.
* fix(server): handle GitLab locked merge state and custom-port glab host keys
- Pass the bare hostname to glab auth status and glab api --hostname (matching ForgeResolver.resolveHost and glab's api_host config) while preserving host:port in the canonical URL and description.
- Map GitLab's transient locked merge-in-progress state to open so watches keep polling until merged or closed.
* fix(server): preserve accurate GitLab watch updates and offline unsubscribe
* fix(server): validate GitLab watch project addresses and retain approval updates
* fix(server): honor GitLab watch profiles written by glab login
* fix(server): refresh authenticated GitLab watch profiles
* fix(server): keep GitLab poll access checks on merge requests
* fix(server): use GitLab MR access and enforce host pauses on every request
* fix(server): let admitted polls finish and retry denied GitLab reads
* fix(server): give quota-limited change request watches a turn
* fix(server): recover from shared forge quotas one poll at a time
* fix(server): rotate watches after failed quota recovery attempts
Keep one titlebar toggle mounted while Explorer opens and closes. Reserve its footprint in the active rail so native pointer clicks and the header tab sequence work, and verify clearance from the Windows window controls.
* feat(plugin): let plugin navigation focus the destination host in the sidebar
Plugins that link across hosts open agents and workspaces with
navigation.openAgent and navigation.openWorkspace, but the sidebar keeps its
host, project and label filters, so the destination can be hidden. Plugins
can't change those client-side filters.
Add an opt-in focusHost flag. With focusHost: true the client selects only the
destination host in the sidebar and clears project and label filters that
could hide it, keeping the grouping mode. Calls without the flag are
unchanged, and navigation.supportsFocusHost lets plugins detect support.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Normalize plugin navigation host IDs
* Preserve sidebar filters for removed hosts
* Focus the host returned by plugin navigation
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mohamed Boudra <boudra.moha@gmail.com>
* feat(skills): install bundled orchestration skills for Antigravity
Antigravity CLI (`agy`), App, and IDE do not scan `~/.agents/skills` or
`~/.claude/skills` globally; they discover global skills one level deep
from `~/.gemini/config/skills/<skill>/SKILL.md`. Gemini CLI already
discovers `~/.agents/skills`, so targeting `~/.gemini/config/skills`
only (instead of `~/.gemini/skills`) avoids duplicate skill entries in
both Gemini CLI and `agy`.
Appends `antigravityDir` at root index 3 in skill transactions so any
in-flight pre-upgrade transaction manifest retains its existing backup
indices, and includes `antigravityDir` in legacy Codex alias checks so
symlinked roots are not retired.
* fix(skills): unify install roots and keep rollback/updates resilient to unwritable targets
- Replace separate per-target fields on SkillTargets with an ordered
`installDirs` list plus `codexDir`, deriving status, sync, removal,
legacy Codex alias checks, and transaction root indexing from one
source while preserving pre-upgrade backup indices 0–2 (`codexDir` at
index 2, additional install roots at index 3+).
- Skip unreadable `add` entries with no backup during transaction
rollback so an ENOTDIR/EACCES parent under `~/.gemini` does not fail
rollback and leave a stuck transaction blocking later skills calls.
- Ignore `ENOTDIR` in `removeSkill` (a skill cannot exist below a
regular file).
- Continue syncing remaining skills and install roots in `syncSkills`
before throwing the first error so an unwritable `~/.gemini` does not
block startup auto-updates from reaching `~/.agents` and `~/.claude`.
* fix(skills): finish removing reachable skills during uninstall when one root is unreadable
Continue across every target root in `removeSkill` and every managed
skill in `uninstallSkills` before throwing the first error (matching
`syncSkills`), while still reporting `EACCES` rather than ignoring it.
* test(agent): synchronize stale-session retry before checking cleared run
Replace the racy `setTimeout(0)` in `agent-manager.test.ts` with a
deferred gate on `resumeSession`. Under CI load, `setTimeout(0)` could
fire after the immediate retry had already resumed and started its new
in-flight run, causing `hasInFlightRun(agent.id)` to read `true`.
* fix(commands): deduplicate provider slash commands by name
Antigravity ACP emits both its built-in slash commands and discovered skills in one available_commands_update without deduplicating colliding names (such as a local plan skill alongside built-in /plan). Both entries flowed through plugin ACP, plugin-provider, and mergeSlashCommandSources into Autocomplete's FlatList, triggering React's duplicate key error on .$plan.
Enforce unique command names (first-wins) at the plugin ACP, plugin provider, core ACP, and Codex boundaries so every daemon consumer gets a valid catalog, and claim provider command names in mergeSlashCommandSources and inline skill filtering so the app stays safe against older daemons.
* refactor(commands): enforce unique command names at daemon and app boundaries
Move slash-command deduplication from six per-provider and per-consumer copies to the two single boundaries: handleListCommandsRequest in packages/server/src/server/session.ts (covering live agents and drafts for every provider) and fetchAgentCommands in packages/app/src/agent-commands/use-query.ts (covering all app consumers, including against older daemons).
* chore(app): tag app-side command deduplication as a COMPAT shim
* fix(app): keep composer and chat column on the outline rail
Previously only WebStreamViewport reserved the 44px chat outline gutter and clamped right gutter on wide web panes (>= 918px), leaving the composer, floating track bar, and callouts at symmetric 16px padding. On Full width, on Wide (1200px) below 1320px pane width, and on Default (820px) between 918px and 939px, the composer extended off the timeline's vertical rail.
Move outline gutter ownership to a platform-split chat-rail module mounted at the chat pane surface. On web it measures the pane width once against the 918px threshold and publishes the shared left and clamped right gutters via CSS custom properties so composer, tracks, and callouts stay aligned with zero resize re-renders or dynamic Unistyles rules, while native stays a zero-hook pass-through. Removing direct outline measurement from AgentStreamView also stops ProviderSubagentPanel from reserving space for an outline it never renders.
* fix(app): keep outline prompt index across 918px crossings and exclude draft error from fullscreen rail
* fix(usage): keep window shorthands visible in sidebar usage row
Antigravity usage windows emitted full group names and raw period strings (for example "Gemini Models weekly") in shortLabel, which pushed the measured labels row past the maximum sidebar width. On the app side, choosePinnedUsageLayout dropped every window on the row straight to bare percentages whenever any scoped short label overflowed the row.
Normalize Antigravity shortLabel periods (weekly -> wk, monthly -> mo) and shorten group prefixes, and add a compact period-shorthand tier between full labels and bare percentages in PinnedUsageTrigger so tight sidebar rows keep shorthands like 5h and wk visible instead of collapsing to percentages alone.
* test(usage): expect compact shorthands in usage modal sidebar summary
With the compact period-shorthand tier between full labels and bare percentages, three Weekly windows now render as "31% wk", "20% wk", "12% wk" in the 320px sidebar instead of dropping to bare percentages.
Replace the website's legacy Paseo mark and plain-text headers with the canonical Paseo butterfly and Paseo Hub flower marks and Plus Jakarta Sans SemiBold lockups from assets/brand/.
Sourcing the website SVG components directly from assets/brand/ keeps mark proportions, 94% butterfly optical balance, and outlined wordmarks aligned with the canonical assets without shipping an extra webfont or duplicating paths across components.
Moves goose into a built-in opt-in provider plugin (`plugins/goose-provider`) and lists models from every provider configured in goose.
- Bundles `plugins/goose-provider` with `server_info.features.gooseProvider`, legacy host catalog fallback, legacy `extends: "acp"` config migration, and legacy session handle decoding (`COMPAT(gooseProvider)`).
- Adds `context.request(method, params)` and synchronous `config(state, context)` hooks to `@getpaseo/plugin/server/acp`.
- Queries goose's provider inventory RPC during discovery to expose `<provider>/<model>` entries (falling back to the base catalog if unavailable), switches provider before model on selection, and namespaces session config snapshots.
Co-authored-by: James Dumay <james@generalinference.xyz>
* feat(antigravity): show account usage
Resolve the provider launch for usage discovery and report quota windows from agy.
Co-authored-by: RoundRedDot <roundreddot@gmail.com>
* fix(usage): clear reports when a provider executable is missing
Usage discovery treated a missing executable as a transient error and kept the previous account report. Treat the unresolved launch as absent so the next refresh removes that report.
---------
Co-authored-by: RoundRedDot <roundreddot@gmail.com>
Co-authored-by: Mohamed Boudra <boudra.moha@gmail.com>
The general GitHub search can fill its result limit with public repositories before returning a private match. Fetch private matches separately so users can find them by name.
Use the native mouse click count to select words and source lines in canvas diffs. Extend the selection by whole words or lines while dragging.
Closes#10116.
* fix(usage): show the current Claude account after a login switch
Agent usage discovery kept the old account bound to a credential store after an account switch. Refresh those discoveries and have the usage source verify account ownership when it rereads the store.
* test(server): synchronize the provider update check fixture
The Windows runner can finish reading the installed version before the update callback starts. Wait for the callback to start and keep it pending while verifying the independent reads.
* fix(app): leave providers that are not installed out of the model picker
An agent's own provider stays in its selector, labelled Unavailable.
* feat(app): open a provider's detail as a Settings page or a sheet
Settings → Providers now pushes a provider page, like Projects and Plugins;
the model picker gear and setup keep a sheet. Both frame one body, ordered
error, Enabled, Installation, Models, Troubleshooting, Danger zone, so the
old sheet's sticky footer and Retry button are gone. Refresh lives once, in
the Troubleshooting status row.
The page header (back link, icon, title) moves into the shared settings kit
so projects, plugins and providers line up on the same rails.
* feat(app): show each provider's version, updates and faults in the list
Each row now says how its provider is: version and model count, Update
available, a spinner while an update runs, a red dot with the error, or
Not installed / Disabled. Only a broken provider gets a dot, and a
provider that is not installed has no switch, since the model picker
already leaves it out. The kebab is gone; Remove lives on the detail.
Rows read the same installation and update queries as the detail, so
opening a provider costs no new requests. Hosts without
providerInstallation still get a correct list from the snapshot alone.
Add provider is a "+ Provider" button in the header and a muted last row;
both open one sheet with search and catalog rows, replacing the tile grid.
After Add, the new provider's detail opens.
The row is the settings kit's detail row, which gains a leading icon, a
hint status (spinner, error dot, or a held line) and a control before the
chevron. The Unistyles test stub now applies withUnistyles theme mappings,
so browser tests render components that read colours through the theme.
* feat(app): open setup's provider pages over the setup card
The detail and Add provider sheets take the setup card's frame and a back
arrow, so they read as pages of setup instead of a sheet over it. The step
is renamed "Your providers" to match the list it shows.
* docs: describe the provider list and pages as they are now
* fix(app): keep settings rows one height and on one icon rail on phones
On touch layouts a switch reserves a 44px control, so in the provider list
rows with a switch stood 10px taller than "Not installed" rows without one.
Card rows now reserve a default control's height, the rule nested rows
already follow.
A compact SettingsRow wrapped with a full `gap`, which added 12px on top of
the leading slot's margin: catalog icons sat 24px from their labels instead
of 12px. The wrap only needs space between lines, so it uses `rowGap`.
* fix(app): keep a provider row's error to one line
A long or multi-line error wrapped the hint to two lines with the dot
centred between them. The row cuts it to one line; the provider's page
shows it in full.
* fix(app): title a provider's Settings page like a project's
The page title and the detail header both named the provider. The screen
now titles it "Providers" and the page names the provider once, the way a
project's page does.
* fix(app): keep a settings row's switch out of the row's button
The provider row's switch rendered inside the row's <button>. That is
invalid markup, and keys reached both: Enter on the switch toggled it and
also opened the provider's page, and Space on it opened the page without
toggling.
SettingsRowHeader now lays its press target under the row's content as a
sibling of the control. The content lets presses through to the button and
stays out of the accessibility tree, where the button's label already names
the row. Hover is tracked on the row, so it holds over the switch.
* fix(app): reserve a control's height only on rows that open something
The previous change gave every card row a control's minimum height and
the design doc said so, but rows built directly on settingsStyles.row
(66 uses) never got it: a single-line row was 48px or 64px depending on
which code path built it.
The minimum now belongs to SettingsRowHeader, the row with an optional
control slot, so a list where only some rows carry a switch keeps one
height. SettingsRow and hand-built rows take their content's height again,
as on the base branch, and the doc states the narrower rule.
* test(app): give the Unistyles stub the light theme's full colours
Since the stub started applying withUnistyles theme mappings, a mapping
reading a colour the stub's theme lacked crashed: the markdown styles read
colors.terminal and failed the plan card disclosure tests. The stub's
colours now start from the light theme, with its own values pinned on top.
* test(app): find the provider page's buttons in the compact sheet
On compact layouts AdaptiveModalSheet puts its testID on the header only,
so buttons in the page body cannot be found inside it. The spec looks them
up on the page, as it did before the page moved into the sheet.
* fix(app): let screen readers read a settings row's state line
The row's button carries the row's name as its label, so only the title is hidden from assistive tech now; the hint below it (a provider's version, error or Not installed) is read again.
The Enter-on-switch test now dispatches its key events synchronously inside one act. An awaited keyboard call let the Switch's animation run a frame after the test, which Vitest reported as an unhandled error and failed the run.
* fix(app): don't open a provider's page after its Add sheet was closed
Adding a provider waits for the daemon to probe it. If the user closed the sheet during that wait, the completed add still navigated to the provider's page. The add now remembers which sheet session started it and only opens the page if that session is still current.
* fix(app): say Not installed for a disabled provider that is not installed
A disabled provider's snapshot status is unavailable, so the row said Disabled even when the binary was missing. The installation inspection is checked first, so Not installed wins over Disabled.
* fix(app): keep a saved provider choice visible when it is not installed
The provider picker drops unavailable providers. A schedule or metadata-generation setting saved against such a provider then showed Select model. The picker now keeps the saved provider, labelled Unavailable, so the user sees what is set and can change it.
* test(app): wait for an added catalog provider's probe before continuing
Adding a provider opens its page only after the daemon probes it, which can outlast the default expect timeout on CI.
* refactor(app): name the provider picker's connection and saved-choice options
Callers passed a bare true to reach the saved-choice argument.
* fix(app): show a provider page's loading and not-found states
A provider page opened from its URL (a reload, a cold start, a shared link) rendered its title over nothing until the host's providers arrived, and forever when the id named a provider the host doesn't have. The page body now says Loading..., that the host is offline, that the providers couldn't be read, or that the provider isn't on this host, in the same notice card the provider list uses when it is offline. The page and both sheet frames share the body.
* fix(app): don't call a just-removed provider missing while its page closes
The daemon can publish the provider list without the removed provider before it answers the config patch, so the page said "This provider isn't on this host" until the answer arrived and it navigated away; in a sheet, through the exit animation as well. A page that has shown its provider now shows nothing once it leaves the list. The not-found notice is for a page that never found its provider. The removal e2e holds the config answer so it always sees that order.
* fix(app): show the not-found notice when a provider is removed elsewhere
The page hid the notice for any provider it had shown before, to cover its own Remove. A removal from another client, the CLI or a plugin then left the page blank with no way to tell why. The page now owns its Remove: while that runs and once it succeeds the body shows nothing, and any other absence says the provider isn't on this host.
The removal e2e routes only the e2e daemon's socket, so the fixture's port-6767 guard stays in force.
* fix(app): show progress on the provider page's in-flight buttons
Remove, Add model and the diagnostic re-run now show their work on the
pressed control, like the page's Refresh and Update: a spinner, no
further presses and, on a labelled button, "Removing..." or "Adding...".
Remove kept its label and only disabled itself, and the page went blank
mid-removal: the daemon publishes the provider list without the removed
provider before it answers the config patch, so the body had nothing to
show and each frame's title fell back to the raw id. The removal now
keeps the entry the user confirmed, and one hook feeds the frame's title
and the body, so both keep showing the provider until the page closes.
The in-flight state also starts after confirmation, not while the
dialog is open.
The diagnostic sheet's header icon buttons had lost the base branch's
spinner and dimming; SheetHeaderIconButton now takes loading and
disabled, the way Button does.
* fix(app): don't leave a re-added provider's sheet stuck on Removing...
The setup step and the model picker's gear keep a provider's sheet
mounted after it closes, keyed by provider. A finished removal stayed
set in that sheet, so adding the same provider back reopened it with
Remove spinning and disabled. A finished removal now ends when the host
lists the provider again.
* fix(app): keep the composer's remembered provider in its picker
The picker leaves out providers that aren't installed, except the one
holding the picker's current choice. The schedule form and metadata page
passed that choice; the new-agent composer didn't, so a remembered
provider not installed on this host vanished from the picker and the
trigger read "Select model" with no reason given. The builder now requires
every caller to say what its current choice is, or null.
* fix(app): guide a missing catalog provider's install and keep custom models while one fails
A provider added from the Add provider catalog runs through the generic ACP client, which reports
no installation, so its page said nothing when the CLI was missing. The page now shows the
Installation section's Not installed row with Install opening the catalog's guide. A link the
host supplies still wins.
Custom models live in config, so the page lists them and offers Add model whenever the provider
is enabled and installed, including while it fails. The provider's own models, and the search
over them, still show only while it runs.
* fix(app): make settings rows pressable on iOS and Android
React Native 0.86 no longer exports StyleSheet.absoluteFillObject, and
Unistyles' native StyleSheet copies it from React Native, so the spread
was empty there. The row's press target then had no size: on Android,
tapping a provider row or a Voice row did nothing. Web was unaffected
because Unistyles' web StyleSheet still defines the key. Use
StyleSheet.absoluteFill, as the rest of the app does.
* fix(app): let setup's provider sheets leave out the close button again
The base's sheet header rewrite dropped SheetHeaderView's showCloseButton,
which the provider page and Add provider sheets pass when setup's back
arrow is the only way out. Restore it on the header.
* test(app): prove a refresh clears a failed update from the provider's row
The update-failure journey now also walks the provider list: after Refresh
on the page the row reads its version and model count with no error dot,
and a refresh from another client clears a row that is open in the list.
Without the provider.installation.update.cleared handling both row checks
fail. The helper's Refresh now uses the page's Troubleshooting button,
since Settings opens the provider as a page, not a sheet.
* fix(app): show a catalog provider's install guide on hosts that cannot report installations
A host without providerInstallation returns an unsupported installation,
which the page passed through, so a catalog provider whose CLI is missing
showed no Installation section and no guide anywhere. The guide is the
app's own catalog data, so the page builds the same Not installed row
from it on those hosts.
* fix(app): apply content width presets to open chats
Retained rows kept the numeric width in their generated web styles. Use a CSS variable so the timeline, composer, and Markdown previews follow the same preference while mounted. Reserve the outline gutter when the outline is enabled on a wide pane.
* fix(app): keep capped chats centered beside the outline
Reserve the left outline gutter and match it on the right when the selected column fits between both gutters.
* feat(providers): show installed version and install link in provider settings
Providers can now declare `installation`: install and update links plus an
`inspect` callback that reports installed state and an opaque version string.
Built-in clients and plugin providers share one contract; plugins receive the
daemon-resolved launch with `resolved: false` when the executable is missing,
so they can report not installed instead of failing.
The daemon calls `inspect` only on the new
`provider.installation.inspect.request` RPC, never from availability probing,
discovery or session creation, so a slow version probe cannot delay agents.
Results are cached per client for 30s, deadlined at 20s, and links survive a
failed probe. The app gates the provider sheet's Installation row on
`server_info.features.providerInstallation`, so older hosts show nothing new.
* feat(providers): show available updates in provider settings
Providers can now declare `installation.checkForUpdates`, returning
`current`, `available` with opaque `versions` and `message`, or `unknown`.
The provider decides what counts as newer; Paseo shows its answer as
written and never parses or compares versions.
The check is its own RPC, `provider.installation.check_updates`, so the
installed version never waits on a network lookup. The daemon caches a
conclusive answer for an hour and an inconclusive one for 30s, shares
in-flight calls, and deadlines the call at 25s, below the plugin RPC
timeout. A throw or timeout comes back as `error`, which the app shows as
inconclusive, never as up to date. The sheet's Refresh bypasses the cache.
Codex and Claude Code read the release source their own updaters use:
the Homebrew cask for cask installs, otherwise the GitHub latest release
for Codex and the configured `autoUpdatesChannel` for Claude Code.
`codex doctor --json` reports the same answer but scans local state and
took close to a minute, so it is not used. Muse compares its build with
the launcher's channel manifest, the same rule the launcher updates by.
The install link now shows whenever the provider declares `installUrl`
and is not reported installed: the Install button for `not_installed`, a
quieter "Install instructions" link when the check failed or the
provider only links to a guide.
* feat(providers): run provider updates from provider settings
Providers can declare `update(context)`, and the Installation section shows an Update
button when an update is available. The daemon owns the operation: one per provider,
joined by later requests, kept in memory past client disconnects, streamed to subscribers
through an owned subscription, and aborted only when the daemon stops. The provider's
answer is the outcome; the daemon adds no verification. After success it drops the cached
inspection and update check and refreshes discovery, without restarting agents.
Codex and Claude run their own `update` subcommand and pass the output through. Muse has
no update subcommand, so it does not declare one.
Updater output is capped at 64 KiB per update so a chatty updater cannot grow daemon memory.
Starting an update requires daemon.manage because it runs commands on the host.
* feat(providers): report installation for the remaining providers
Copilot, OpenCode, Pi, Oh My Pi, Antigravity and Hoplite now show an
install link and the installed version in provider settings. Each one
declares only what its CLI offers:
- Update checks ask the source the CLI's own updater reads: Pi's
latest-version endpoint, `omp update --check`, and Hoplite's
latest.json. Copilot and OpenCode resolve their release inside the
binary, so they declare no check.
- Updates run the CLI's own command: `copilot update`, `opencode
upgrade`, `pi update --self` (leaves Pi packages alone), `omp update`.
Antigravity and Hoplite document no self-update.
OpenCode's success message asks for a daemon restart, because all
OpenCode agents share one server that keeps running the old version.
* docs(providers): document the provider installation contract
Plugin authors get the contract in the provider guide, with a worked
example that reports a version, checks for an update and runs the CLI's
own updater, and every field, deadline and cache in the reference.
The internal provider and plugin docs record why the three callbacks are
separate, that versions are opaque, that Paseo adds no verification to
an update, and how the plugin bridge carries and cancels each request.
* chore(protocol): mark provider installation as shipping in v0.17.0
main is on the 0.16 beta series, and a promotion ships the tested beta's
content without pulling from main, so this lands in the next minor.
* fix(app): bound update output and show refresh failures in provider settings
The screen appended every progress chunk with no limit while open, so a
noisy updater grew memory and never showed the truncation note that a
reopened screen shows. Progress now folds with the daemon's 64 KiB limit.
A manual refresh stored its result and failure outside the queries the
screen reads, so a failed refresh brought back the old answer with no
warning. Refresh now runs through those queries. It passes staleTime 0
because the app's default staleTime of Infinity would otherwise answer
from the cache.
The not-connected error uses the translated string the i18n check
requires.
* fix(providers): keep update checks in step with the installed version
A conclusive check is cached for an hour and was not tied to the version
it answered for. After a CLI updated outside Paseo, settings showed the
new version and still offered Update to it. When an inspection reports a
different version, the cached check is dropped. Versions are compared
for equality, so they stay opaque.
The release comparison dropped prerelease suffixes, so Codex
0.200.0-alpha.1 looked current against 0.200.0. A prerelease now comes
before the release with the same numbers. Two different prereleases of
the same release are reported as unknown.
* docs(providers): say when Update instructions shows
* fix(providers): reuse a cached update check only for the version it answered
The previous fix dropped the cached check after an inspection settled
with a new version. On reopen the app sends inspect and check together,
so the check answered from the stale entry before the inspection
settled. A cached check now records the installed version it was
computed against and is reused only when the current inspection, shared
if in flight, reports the same version.
* fix(app): refresh provider version after an update settles while disconnected
The installation and update queries were invalidated only by a live
succeeded progress event. A connection dropped mid-update delivered the
settled operation as a snapshot on reconnect, which left the old version
and Update on screen. A success the view saw running now rereads them
whether it arrives as progress or as a snapshot.
* fix(providers): keep update checks within their own deadline
Tying a cached check to its version made every check wait for a fresh
inspection, which could add the 20 s inspect deadline to the 25 s check
deadline. A check now never starts an inspection. It waits on one only
when the inspection is cached or already running and the check would
reuse a cached answer. When it asks the provider, it records the version
afterwards.
* test(app): assert provider installation states are visible
* fix(providers): share one update check when two clients reopen after a version change
* fix(muse): report the build the update check compares
Paseo reuses a cached update check while inspect reports the same
version string. Muse reported 1.4.18 but compared its build
(1.4.18-R8600.1), so a new build of the same release kept the stale
"Update available" for up to an hour. The contract docs now say to
include everything the check compares in the reported version.
* test(app): scroll custom model rows into view before checking them
The provider sheet now shows an Installation section above the model
lists. At 390x1000 the custom model row sits under the sheet footer
until scrolled. The spec checks that a long model id leaves its Remove
button reachable. It had assumed the row needed no vertical scroll.
* test(app): keep the custom model check sensitive to a Remove button pushed off the sheet
scrollIntoViewIfNeeded on the Remove button scrolled the overflow-hidden model
card sideways too, so the check passed with the long-id layout fix undone.
Scroll only the Custom models heading into view, which moves the sheet
vertically and leaves the card at scrollLeft 0. With the layout fix undone the
spec now fails at 1200px and 390px again.
* feat(app): offer provider updates whether or not an update check answers
Update and Update instructions lived inside the row that appeared only after
checkForUpdates answered, so Copilot and OpenCode, which declare update but no
check, never offered either. A check is a convenience: the Updates row now
shows for an installed provider that offers an update, keeps Update or Update
instructions in place in every check state (none, checking, up to date,
available, couldn't check), and shows what the check knows as a hint beside
it. A provider that offers no update still shows the check's answer once it
has one. The daemon already ran update without a check; a test now says so.
* fix(app): let update output scroll inside the provider sheet on Android
The output box is a ScrollView nested in the sheet's scrolling body. Android hands the drag to the outer ScrollView unless the inner one sets nestedScrollEnabled, so only the last 240px of a long update's output was readable. The tool-call detail views set the same flag.
* fix(app): show one progress indicator per provider update and refresh
While an update ran, the Update button and the update's row each showed a spinner. The button's
spinner also widened it, and the row's leading spinner shifted the title when the update settled.
The update's row now owns progress from the tap until the update settles: it shows the update
starting before the host answers, so the indicator never moves, and the Update button only
disables.
Refresh blanked the Installation rows while the provider was asked again, then moved "Checking…"
from the Version row to the Updates row. The rows now keep their last answers until the new ones
arrive, and the Refresh button shows the refresh until both the models and the installation are
read again.
* fix(app): show a provider update's progress on its Update button
The pressed button carries an action's progress everywhere else in Settings: Button's own
loading state with a present-participle label, as Browser data's Clear and Project settings'
Save do. Update now does the same, reading "Updating..." from the tap until the update settles,
and the update's row holds only the output while it runs, then the result. This replaces the
row-owned spinner from the previous commit, which no other surface uses.
The installation's in-flight labels use three dots like their siblings.
* fix(app): keep a running update's progress visible whatever the inspection reads
Update carries the update's progress, but its row rendered only while the inspection read
"installed". An inspection that ran during the update (on reopen or Refresh) could read loading,
not installed or failed while the binary was replaced, and the sheet then showed nothing about the
running update. The Updates row now stays while an update runs.
* fix(server): show and update the provider CLI its agents launch
The installation contract found the CLI on the daemon's PATH, while agents
spawn the bare command with the provider's environment. A provider whose env
sets PATH could launch one copy and show and update another, so an update
succeeded while new agents kept the old version. The CLI is now resolved with
the provider's environment.
The section test's check-state table now always names the text it expects, so
no case can skip its assertion.
* test(server): name the provider PATH test for what it shows
The availability check that gates a Copilot spawn still looks on the daemon's
PATH, so the copy on the provider's PATH is not always the one agents launch.
* test(antigravity-provider): read the version from the CLI, not its acp subcommand
Antigravity now launches `agy acp --disable-telemetry`, so the launch the
installation receives ends with the ACP subcommand. The version read drops it,
since `agy acp --version` starts the ACP server.
* feat(plugin): pass the inspected version to checkForUpdates
checkForUpdates now receives installedVersion, the version inspect reported, and
the daemon keys the cached answer on it. Before, each provider read its version
a second time and the daemon recorded which version a cached check answered
for, so a check could compare a different version than the one shown, and a
plugin had to put everything its check compared into the reported version.
An update check now waits for the provider's inspection (shared with a
concurrent inspect, refreshed with it), answers Not installed or the
inspection's error without asking the provider, and is asked again once the
inspection reports a different version.
* fix(server): clear a failed provider update when the provider is refreshed
The daemon kept a provider's last update for its whole lifetime, so a failed
update stayed the current update until a later update succeeded or the daemon
restarted, and the provider looked broken after the user fixed it. A refresh
of the provider's installation or update check now drops a failed result. A
running update is never dropped.
* docs(server): say which agents look up the provider CLI on its PATH
* fix(server): tell open provider views when a refresh drops a failed update
A refresh deleted the failed update on the daemon but sent nothing, so a view
already observing the provider's updates, in this client or another one, kept
showing the failure until it observed again. The daemon now pushes
provider.installation.update.cleared with the dropped operation's id to every
subscriber, and the app clears the update it shows when the ids match.
When a worktree folder disappears, the workspace git service can cache a
not-git snapshot before it notices and suspends the workspace. After the
folder comes back, a plain getSnapshot returned that cached snapshot
before it tried to resume, so checkout status said "not a git repository"
until the 60 s missing-folder probe. The Changes panel of a just-restored
worktree had no repository header during that time.
Resume a suspended workspace first, then serve the cached snapshot.
This made e2e worktree-restore.spec.ts:167 flaky in CI.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(agents): keep waiting heartbeats out of Running
* fix(protocol): keep the work count default as a wire contract
* test(app): verify heartbeat status before, during, and after its turn
* fix(server): notify once while an agent's finish is unread
Since v0.16.0, a turn that an agent starts on its own cleared the
agent's unread "finished" attention. Every such turn that then ended
raised a new "finished" notification. A Claude Code agent that reacts
to each event of a Monitor or a background task starts one turn per
event, so the user got a notification every few seconds.
A turn that the agent starts on its own now clears only error
attention, which is what the earlier change intended. An unread finish
stays unread, so the next finish does not notify again. After the
user reads the agent, the next finish notifies as before. A prompt
from the user still clears all attention.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(server): notify a repeat finish the user was not told about
The previous commit kept an unread finish across turns that an agent
starts on its own, and the agent manager then sent no attention for
the later finishes. Suppose the user had the agent on screen for the
first finish. The app keeps that finish unread, and no notification
is sent while the agent is on screen. After the user left, no later
finish notified at all.
The agent manager now reports each finish, and marks a finish that
repeats an unread finish. The websocket server owns delivery, so it
records whether it delivered a notification for the unread finish.
A repeat notifies only when nothing was delivered for it yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* fix(server): key the notified unread finish by its attention timestamp
The websocket server recorded only that it had notified for some
finish of an agent. Mark as unread starts a new unread finish without
an attention event, so a later repeat used the record of an older
finish that the user had already read, and sent nothing.
The attention callback now carries the timestamp of the agent's
current unread attention, which identifies the unread finish. The
server stores the timestamp it notified for, and suppresses a finish
only when it belongs to that same unread finish. This replaces the
repeatsUnreadFinish flag.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(server): drain starting file observations during shutdown
Workspace Git aborts starting watchers on shutdown. The observer only tracked completed subscriptions, so close could return while a startup scan was still releasing its resources.
* test(server): verify watcher release during startup shutdown
Use an internal backend adapter to hold startup open and observe watcher release when close returns.
* fix(server): leave startup scan drainage to watcher backends
Both backends already drain their admitted scans. Preserve cleanup for a failed root lookup after the observer takes ownership up front.
Pressing the model picker shortcut now focuses search without opening the button tooltip. Attach picker hints to the shared trigger button so focus inside the picker stays independent.
* Keep a file view's scroll offset when switching Preview and Source
A file tab rendered either its Preview or its Source, so each switch
unmounted the view on screen and its scroll offset with it. Coming back
started at the top of the file.
Each view now stays mounted once shown and is hidden with RetainedPanel
while the other view shows. The hidden Preview keeps the text it last
showed, so editing in Source does not render it on every change.
Co-authored-by: apple-ouyang <45086632+apple-ouyang@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Assert what the file pane shows, not that the other view unmounted
The Preview and Source views now stay mounted once shown, so the file
editing spec checks the visible heading and a hidden HTML preview.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Load the HTML preview again in the WebView that shows after a hidden panel
iOS destroys the native WebView of a hidden retained panel and makes a new
one when the panel shows. The navigation guard latched the document per
component mount, so it refused the new WebView's load of the same document
and the preview stayed blank after Source -> Preview. The guard now releases
the latch while its retained panel is hidden.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Name the iOS HTML preview as the view that does not keep its offset
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Load a changed HTML document in a new preview frame on web
The Preview stays mounted and hidden while Source shows. When the file
changed meanwhile, the Preview got the new document in the same render
that showed it again. Chromium left the reused iframe blank in that case
until the next Source/Preview switch. Key the iframe by its document so
each document loads in a new frame.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Co-authored-by: apple-ouyang <45086632+apple-ouyang@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(server): branch off a pull request and list every git worktree
A workspace.create worktree source with action "branch-off" and a
checkoutSource used to check the pull request out, because core took
any change request as a checkout. Branch-off now fetches the pull
request's head into a remote-tracking ref (origin/<head> for a
same-repo head, paseo-pr-<N>/<head> otherwise) and starts the new
branch there, with no upstream and no remote added. A request with no
action still checks the pull request out, as older clients expect.
checkout.worktrees.list returns all git worktrees of the repository,
not only the ones Paseo created, so the New workspace screen can offer
any of them. Bare and prunable entries are left out.
Hosts advertise both through server_info.features: worktreeRefActions
and checkoutWorktreeList. Older hosts ignore the branch-off action on a
pull request, so the app must gate on the flag.
* feat(app): pick an existing worktree, and branch off or check out a ref
Isolation now offers Project directory (was "Local"), New worktree, and
Existing worktree. Existing worktree is a local workspace at the chosen
worktree's path, not a third isolation value, so plugins, preferences, and
the create request still see "local". The list comes from
checkout.worktrees.list, so it includes worktrees Paseo did not create.
The ref picker gains Branches / Pull requests pills and a Branch off /
Check out toggle with a tooltip per action. Branches default to Branch off
and pull requests to Check out. On a host without worktreeRefActions the
toggle is hidden and the defaults apply.
Git keeps a branch in one worktree at a time. Picking a held branch in
Check out mode moves to the worktree that holds it, and creating from a
held branch that was never picked fails with an explanation. Without
this the server would silently check out a "<branch>-1" copy.
* docs: describe Existing worktree and Branch off / Check out
* fix(server): give a fork pull request branch-off a base the repository has
A branch off a fork's pull request stored the fork's head as its base, under
refs/remotes/paseo-pr-<N>. Create PR then asked for a pull request into a
branch the base repository does not have, and the ref piled up in base branch
pickers because no remote owns it. The new branch now starts at the fetched
head, kept under refs/paseo/change-requests/<N>, and its base is the pull
request's base branch. A same-repository branch-off still stacks on the pull
request's branch.
A routed create_workspace with a prNumber and no mode checks the pull request
out again, as before branch-off could start from a pull request.
The worktree list skips a locked worktree whose directory is gone, since it
cannot be opened.
* fix(app): start an existing worktree's agent in that worktree
Existing worktree created the workspace at the worktree, but the composer and
terminal still used the project directory, so the daemon rejected the first
agent as outside the workspace. The screen now derives one working directory
from the isolation and uses it for the workspace, the composer, and the
terminal. A worktree that left the list is no longer a choice.
The ref picker names change requests the way the project's forge does, so a
GitLab project shows Merge requests. The schedule form uses the isolation
labels Project directory and New worktree.
* fix(server): compare a fork pull request branch-off against the remote base
The base was stored as the bare branch name, which resolves to the local branch
first. A local main behind origin/main put upstream commits in the workspace's
changes for its whole life, and a base branch the clone never fetched failed
creation. The base branch is now fetched from origin and stored as its
remote-tracking ref.
* fix(server): take a pull request's base and a checked-out branch from the remote that has them
A fork branch-off fetched its base from origin. In a clone whose base
repository is upstream, origin is the user's fork or missing, so the base was
stale or the create failed. The base now comes from the remote that served the
pull request, and a failed base fetch falls back to the bare branch name.
Check out of a branch that only a remote other than origin has failed with an
unknown branch. The local branch is now created from the one remote-tracking
ref with that name, tracking it, as git does.
The fork branch-off test passes a git identity to commit-tree, which CI
runners do not have.
* fix(server): branch off a pull request from the remote that serves it
A same-repository pull request's head was fetched into refs/remotes/origin.
In a fork clone, upstream serves the pull request and origin is the fork, so
the base refresh replaced the head with the fork's branch of the same name and
the new branch started from code that is not the pull request. The head now
lands under refs/paseo/change-requests/<N>, the branch starts there, and the
base is fetched from the remote that served the head.
A routed create_workspace that branches off a pull request is refused, as it
is on the local host.
* fix(app): branch off a plugin's base ref whatever the branch action
setBaseRef chose the base but kept the user's Check out switch, so a plugin's base became a checkout or a held-branch error. A base is what a new worktree starts from, so setBaseRef switches branches back to Branch off.
Four worker-terminal-manager tests SIGKILL a terminal worker whose
terminal program runs in the test's temporary home. On Windows that
program ends only when the dead worker's console closes, a moment
later. afterEach then removed the home with a 0.5 s retry budget, and
Windows refuses to remove a directory that is a live process's working
directory, so cleanup failed with EBUSY.
The tests now kill a worker through crashWorker, which records the
worker and its terminal programs. afterEach waits up to 10 s for them
to end before it removes the temporary dirs.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(antigravity): run sessions through the shared ACP provider
Use Antigravity’s ACP server for workspace sessions, mode changes, approvals, and conversation history. Keep existing conversations importable and migrate saved Full access sessions to Default.
* fix(acp): finish delayed history replay before sessions become ready
Keep delayed load updates on the history side of the readiness boundary. Reuse the ACP catalog and configuration mapping to translate legacy Antigravity selections and document the approval requirements for scheduled runs.
* fix(antigravity): preserve Full access in ACP sessions
Keep unattended Antigravity runs working through the shared ACP permission policy. Commit a Full access change only when the session is idle and retire unanswered approvals when a turn ends.
* fix(acp): cancel pending approvals when stopping a turn
Send the approval cancellation when the client asks the agent to stop. Agents that wait for the approval response can then finish the turn, and late replies to closed approvals leave the session usable.
* fix(server): keep Changes available when Git metadata watching fails
A valid checkout could lose its first comparison when metadata observation exceeded the directory limit. Keep working-tree observation active and share a backing-off metadata sample per Git root so quiet comparisons stay idle.
* fix(server): probe only revisions used by degraded Changes streams
Reuse each comparison's initial Git resolution as the baseline. Batched ref lookups run after subscription setup, stay bounded when the repository has a large ref table, and refresh only comparisons with changed inputs.
* fix(server): report Changes metadata fallback in daemon diagnostics
Log the watcher failure once per root and count comparisons currently using the fallback. Exercise the transition from an active watcher to polling and the release of the last comparison.
* fix(server): keep Changes probes working across Git versions and worktrees
Use line-delimited batch input for older Git versions. Resolve shared refs in the common Git directory so a removed worktree cannot block the remaining comparisons.
* fix(server): isolate invalid revisions in Changes polling
Git aborts a batch when an upstream or reflog expression cannot resolve. Resolve expressions independently so a failed lookup leaves other comparisons current.
* fix(timeline): keep replies visible across idle notifications
Resolve notification membership at the manager's live event boundary. An idle notice gets a display turn without starting a run, and in-flight notices join the open run.
* test(timeline): keep notice paging assertions compatible with async history
* fix(timeline): reuse managed turn assignment for notices
Resolve notices before waiter matching with the same rule used by live persistence. Verify untagged foreground runtime notices through runAgent and idle Pi notices through timeline paging.
* fix(app): acknowledge accepted workspace taps before navigation
Short taps finish before the delayed hold feedback paints, then navigation occupies JS. Paint accepted feedback on the native UI thread while retaining the existing press cancellation and delay rules.
* fix(app): avoid persistent tint on selected docked workspace rows
* perf(app): reduce workspace mount work and narrow panel subscriptions
A tap on the web arrives as pointer events: the finger enters the row as it
lands and leaves as it lifts. The submenu row and its flyout read both as
hover, so the leave queued a close just as the press opened the flyout, and
it disappeared about a quarter of a second later. A tap inside an open
flyout closed it the same way.
Hover intent now ignores pointer events whose pointerType is touch. Mouse
and pen behave as before, including on touch screens with a mouse attached.
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
* fix(browser): let agents choose which desktop opens a tab
Registration order sent new tabs to the wrong desktop window. Let the agent discover and select the intended window by name, and require a choice when several windows can open tabs.
* test(desktop): run browser host selection in a fresh app session
Closing the second window can leave the existing browser journey in a different renderer state. Give both journeys their own windows and daemon so the routing test does not change the old test setup.
* fix(browser): keep window names visible and stable across reconnects
Frameless desktop windows need an in-app place to see and copy their name. Give each window one browser host identity so a reconnect replaces a lingering connection without creating a duplicate choice.
* test(desktop): cover host selection in the browser journey
Open the second window after the existing browser checks finish so their setup remains isolated within one CI run.
* fix(app): keep dictation working after closing another chat
Keep the shared native recorder alive while a client still uses it. Report microphone startup errors at the composer so the user can retry from the same control.
Co-authored-by: Brandon Elliott <923787+Brandon168@users.noreply.github.com>
* fix(app): explain how to recover from dictation startup failures
Classify known microphone failures at the capture boundary. Keep daemon stream errors available to the composer and clear stale capture feedback on the next attempt.
* fix(app): keep permission guidance when the native request fails
* fix(app): show dictation errors after a terminal prompt shortcut
---------
Co-authored-by: Brandon Elliott <923787+Brandon168@users.noreply.github.com>
When Paseo used an OpenCode v2 service it did not start (a discovered
`opencode serve --service`, or a configured server URL), nothing
reported that service going away. The connection's `exited` signal
resolved only when Paseo itself stopped using it. A turn that was
running when the service died was never failed, the event stream
retried the dead port every 100 ms, and Stop, send and reload were
refused with "active run cancellation was not acknowledged".
The shared connection now treats a refused connection as the service
exiting, the same signal an owned server's process exit gives. The
session fails the turn, stops its stream, and reconnects on the next
operation.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(server): let agents watch pull requests
* fix(server): isolate PR watch test homes
* fix(server): preserve watch delivery during subscription changes
* fix(server): batch co-due PR observations and drain admitted updates
* fix(server): settle combined poll requests with await
* fix(server): deliver overdue PR updates together
* fix(server): include failing check links in PR watch updates
* test(server): verify PR watch check links on live GitHub
* fix(server): deliver PR digests through the shared agent queue
* test(server): verify PR and child updates share the next turn
* fix(server): withdraw queued updates before provider admission
* feat(server): show PR watches as agent background work
Render active watches beside heartbeats so users can see and stop them while the agent is idle. Bind Stop to the original subscription so a delayed request leaves a replacement intact.
* fix(app): make agent controls searchable before the first message
New workspace already renders the shared draft controls, but Command Center registration lived in the workspace tab. Register from the draft controls and use the active composer scope so each host gets the same behavior.
Co-authored-by: Christoph Leiter <135764+cleiter@users.noreply.github.com>
* fix(app): keep agent settings searchable in fullscreen
Compact fullscreen editing used to unmount the toolbar and remove the active composer settings from Command Center. Keep the controls registered while fullscreen hides the inline toolbar.
* fix(app): show composer settings from compact sidebar search
The sidebar suspends editor shortcuts while it covers the pane. Give Command Center an explicit target so sidebar search still uses the center composer and Explorer keeps its own scope when open.
* test(app): cover compact Command Center settings for running agents
---------
Co-authored-by: Christoph Leiter <135764+cleiter@users.noreply.github.com>
* fix(codex): offer the Flex speed and thread it through
Speed choices came from the model list's `serviceTiers`, and Codex never
advertises Flex there: `ModelInfo::supports_service_tier` treats it as an API
request option that is valid whatever the model advertises. So the Speed menu
could never list it, choosing `flex` threw "Codex speed 'flex' is not
available", and a stored `service_tier: "flex"` was rewritten to Normal before
the first turn.
Flex now joins Normal as a tier Paseo knows, on every Codex model. The Speed
control and the tier checks read one list, so what the control offers is what
the session runs: `turn/start` and rewind's `thread/fork` carry
`serviceTier: "flex"`, a saved Flex survives open and model switches, and a
catalog that does advertise Flex keeps its own entry instead of duplicating it.
Ref: #9773
* Verify Codex Flex in the browser Speed menu
* Gate unadvertised Flex on Codex 0.156
Codex 0.155 silently strips Flex from outgoing requests when the catalog omits it. Use the connected app-server version to gate the locally added option and normalize unsupported saved selections. Keep the Speed icon highlight for any explicitly selected tier and verify Flex follows that behavior.
---------
Co-authored-by: Mohamed Boudra <boudra.moha@gmail.com>
When a toast replaced another just as the other's timer expired, the
old timer fired before the viewport rescheduled it and dismissed the
new toast. The new toast showed for one frame and faded out. The
timer now dismisses only the toast it was set for.
This made "Reloaded npm-review" vanish in the plugin-management e2e
journey when the reload finished about 2200 ms after the install
toast.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* feat(app): rewind the latest turn from Command Center
Make the latest prompt easy to revise without finding its message menu. Focus the composer when a rewind restores that prompt so editing can resume immediately.
* fix(app): hide rewind for attachment-only prompts
Keep the Command Center and message menu eligibility aligned. Rewinding an attachment-only prompt removes the turn without a prompt the composer can restore.
* fix(app): wait for current history before rewinding the latest turn
Keep cached prompts out of the Command Center rewind target while the active chat catches up. The visible-message menu still targets the specific prompt the user chose.