Reproduces stale compaction references through the current transcript-store path. Native resume of the synthetic source, fork, and repaired control succeeds, so this is not evidence of the reported context overflow.
Co-authored-by: kesiyuan <kesiyuan@bytedance.com>
* Let agents steer a running turn from send_agent_prompt and paseo send
When an agent used the send_agent_prompt MCP tool on a target that was
mid-turn, sendPromptToAgent ran without activeTurnBehavior, so the
prompt path fell through to replaceAgentRun. That cancelled the target's
active turn and every provider subagent it had in flight. The web UI,
child finish notifications and schedules already pass "steer"; the MCP
tool and the CLI had no way to ask for it.
Add an optional activeTurnBehavior input to send_agent_prompt. It
defaults to "steer" for agent-scoped callers and to "interrupt" for
top-level callers, so existing top-level behaviour is unchanged. Add
--steer to paseo send. Document both and cover them with tests.
* fix(agents): retain finish notifications during steering acknowledgement
Observe completion before dispatch and hold delivery until acceptance. A replacement discards the displaced turn observation and starts a fresh listener before its iterator runs.
* fix(agents): preserve notification ownership when follow-ups fail
Keep the previous observation until dispatch succeeds and suppress duplicate delivery during acknowledgement. Decide whether to retain the running turn from the dispatch result, including side commands requested with interrupt.
* fix(agents): deduplicate notifications across overlapping follow-ups
---------
Co-authored-by: Mohamed Boudra <boudra.moha@gmail.com>
Authored Markdown destinations were filtered by the filename heuristics intended for inferred text, silently ignoring screenshot links. Resolve explicit destinations directly through the existing workspace opener so reads and failures reach the file pane.
Preview links bypassed the directory-aware opener and opened folder destinations as file tabs. Use the existing opener and explorer reveal actions, retaining preferred file placement and canceling pending checks when leaving the source tab.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Explicit Markdown hrefs were rejected by the filename heuristic used for inferred text. Let authored nested paths resolve without a recognized extension while preserving inline-code, automatic-link, and domain handling.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
The compact composer used a hardcoded box shadow shared by both themes,
so it read as a heavy band on light backgrounds. It now uses the shared
sm shadow token, whose light value is raised from 0.02 to 0.08 opacity
so the lift stays visible. Dark sm is unchanged.
resolveClaudeCodeVersion() probes <configured command> --version and parsed any output with the generic fallback in parseClaudeCodeVersion. With a wrapper command (e.g. cux), that output is the wrapper's own version, which then failed every minimumClaudeCodeVersion gate and silently hid Opus 5.5, Opus 5, Fable 5, Sonnet 5.5 and Haiku 5.5. Only trust output that identifies itself as Claude Code (x.y.z (Claude Code)); anything else falls into the existing unknown-version path that does not gate models.
* Wait for plugin reload completion instead of the request deadline
The daemon answers plugin.reload.request only once the plugin is running
again or has failed, and a reload queues behind other plugin lifecycle
work such as an install with a preparation command. The shared client
applied its sixty-second session RPC deadline to that request, so a slow
reload failed with "Timeout waiting for message" while the daemon kept
loading the plugin, inviting a reload the daemon was still running.
Reload now waits for the daemon's own answer. Enable and disable keep the
session deadline, and a real startup failure still reaches the caller.
* Cover a slow plugin reload end to end
The CLI fixture now holds its own git build on a gate file, starts a
reload of an installed plugin behind that queued install, observes the
reload request at the proxy, keeps the gate closed past the client's
former one-minute deadline, and then asserts the reload's real result:
CLI exit 0, the plugin running, and exactly one reactivation. Listing the
plugin during the wait stays readable and does not restart it. A plugin
that cannot start still fails its reload with the daemon's error.
The request proxy that recorded install sources becomes a small observer
that records any session request, so the wait is measured from the
request boundary rather than from process start.
Verified: npx tsx packages/cli/tests/e2e/plugin-lifecycle.test.ts exits 0
in 135 s; the same fixture without the new scenario takes 66 s.
* Document how a reload reports its result
The CLI docs now say `paseo plugin reload` waits for the daemon to stop
and start the plugin, that cancelling the command or losing the
connection does not stop the daemon's work, and that `plugin ls` shows
the previous state until the reload finishes.
The plugin reference records the lifecycle fact behind it: a reload
queues behind the daemon's other plugin work, the daemon answers when
the plugin is running again or has failed, and status reads, Settings and
the logs are not the reload's outcome.
* Cover the Settings reload wait end to end
Adds a browser scenario for the ticket's user-facing half: with a gated
git install holding the daemon's plugin lifecycle queue, the row's Reload
goes out, the test keeps the gate closed past sixty seconds, asserts the
row stays pending with no error feedback, then releases the build and
asserts the daemon's plugin.reload.response and the app's "Reloaded
e2e-plugin" result.
The scenario turns plugins on through the daemon instead of the global
switch, whose transient "Plugins enabled" confirmation is not its
subject. The wait is measured from the reload request frame, which is an
outbound frame, not from the response stream.
Verified: npx playwright test --project=browser
e2e/browser/plugin-management.spec.ts -g "keeps Reload pending past a
minute" -> 1 passed (1.5m); plugin-reload-pending.png,
plugin-reload-past-deadline.png and plugin-reload-completed.png show the
pending row, the surface past the former deadline with no error, and the
finished reload.
* Correct the status guidance for a cancelled reload
Review finding: the note said a listing keeps the plugin's previous
state until the reload finishes, but that only holds while the reload is
still queued. Once it executes, the daemon removes the plugin from the
runtime catalog before stopping it (plugins/runtime.ts:377-382) and
resolvePluginStatus maps an enabled plugin that is not running to
"failed" (plugins/index.ts:71-78), so a listing read in that window
shows failed for a healthy reload.
Both passages now separate the two windows and say that neither a
listing nor the logs report the outcome of the outstanding reload.
The interim "failed" itself is TUC-1401.
* Harden the reload coverage against failures and lost connections
Review findings folded into the tests:
The Settings scenario releases its gated build in a finally block. Before
this, a failing assertion left the build polling forever, the daemon's
plugin lifecycle queue stayed blocked, and the fixture teardown's own
removePlugin timed out behind it. It also asserts the row's actions are
enabled again once the reload settles, next to the result feedback.
The connection-close regression now reconnects and asserts nothing
resends the lost reload; the no-replay invariant already covered for
legacy requests now holds for a pending reload too.
Verified: npx vitest run packages/client/src/daemon-client.test.ts -> 161
passed. npx playwright test --project=browser
e2e/browser/plugin-management.spec.ts -g "keeps Reload pending past a
minute" -> 1 passed (1.4m).
* Let the gated build finish before its files go
Review finding on packages/cli/tests/e2e/plugin-lifecycle.test.ts:450:
the finally block released the gate and deleted the gate directory
without waiting for the install and the reload it had started. The build
polls for the gate file, so cleanup could delete the release before the
child saw it and leave the build waiting until daemon shutdown.
The fixture now keeps both promises, releases the gate, and awaits them
before deleting their files.
Verified: npx tsx packages/cli/tests/e2e/plugin-lifecycle.test.ts -> exit
0, 161 s.
* Read the reload scenario as intent, not transport code
Review finding: the new browser scenario kept frame parsing, selectors,
polling, deadline arithmetic, screenshots and gate cleanup in one body.
docs/testing.md asks for complex setup in reusable helpers.
The wire observation, the gated queue hold, the row-state assertions, the
deadline wait and the surface captures are helpers now, and the body
reads as the flow: open Settings, hold the queue, reload, still pending
past the deadline, release, read the daemon's result.
Verified: npx playwright test --project=browser
e2e/browser/plugin-management.spec.ts -g "keeps Reload pending past a
minute" -> 1 passed (1.4m).
* Keep the reload lifecycle fact in one document
Review finding: the cli.md paragraph repeated the queue and interim-state
explanation that the plugin reference already owns. CLAUDE.md: "One fact,
one doc. Every other mention is a link."
The reference keeps the full explanation; the CLI page states what the
command does and links there.
* Release the gated build when its setup fails
Review finding: holdPluginLifecycleQueue started the install, then
waited up to 30 seconds for the build's started marker. A preparation
that took longer made the helper throw before it returned its cleanup
handle and before the test entered its try block, so nothing wrote the
release file: the build polled forever, the daemon's plugin lifecycle
queue stayed blocked, and the fixture's own removePlugin timed out
behind it.
The helper now owns that failure path: it releases the gate and drains
the install before rethrowing, so a queue it hands out is always
settleable.
Verified on Linux (Chromium, Playwright):
- With the started marker suppressed in a temporary local patch, the run
fails at the 30-second marker wait inside the helper and the teardown
still completes (both attempts ~33 seconds, no removePlugin request
timeout).
- With the marker in place, the scenario passes: 1 passed (1.4m).
* Keep an early install failure out of the unhandled path
Review finding: the marker wait in holdPluginLifecycleQueue ran before
the install promise had a rejection handler, so a preparation that failed
or a connection that closed before the build wrote its marker surfaced as
an unhandled rejection, which fails the run before the helper's own catch
can drain the queue.
The helper attaches a handler the moment the install starts and races the
marker wait against the install, so an early failure reaches the caller
as the daemon's own error, a build that never writes its marker still
fails as a 30-second marker timeout, and every path releases the gate and
drains the install.
Verified on Linux (Chromium, Playwright), patching the build command
locally for each failure branch:
- Early failure (build exits 1): fails in 4.2 s with the daemon's
DaemonRpcError for plugin.source.install.request, no unhandled
rejection, teardown completes.
- No marker (build waits for the gate): fails at the 30-second marker
wait, no unhandled rejection, teardown completes with no removePlugin
request timeout.
- Unpatched: 1 passed (1.5m).
* Keep a failed launch from ending the run before its cleanup
* fix(app): open folders from inline path links in the file explorer
Clicking a folder path in an assistant message opened a file tab that
failed with "Requested path is not a file". Extension-less
workspace-relative paths without a line number are now probed with a
directory listing before opening: directories open the Files explorer
navigated to that folder, everything else opens as a file like before.
Lexical directories (trailing slash, workspace root) now navigate the
explorer to the folder instead of only preloading its listing.
* fix(app): reveal linked folders and their contents
Resolve workspace folder links from directory metadata and expand their ancestor paths in Files. Preserve file and line-reference opening, including extensionless files.
Co-authored-by: Moy <moyfoxther@gmail.com>
* fix(app): cancel stale path opens and reveal hidden folders
Keep pending directory checks inside the path-opening model so a later click or navigation supersedes them. Explicitly revealing a hidden directory also enables the existing hidden-file visibility preference.
---------
Co-authored-by: Moy <moyfoxther@gmail.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(server): return actionable OpenCode bridge errors
* fix(server): surface OpenCode tool input errors
* fix(server): surface OpenCode v2 tool errors
* test(server): cover OpenCode v2 bridge errors
* fix(server): classify missing schedules as tool input
* fix(server): surface schedule state conflicts
* fix(server): validate schedule tool inputs
* fix(server): keep tool contracts while reporting OpenCode input errors
Preserve the removal of create_agent.background and keep legacy schedule fields hidden. Verify client errors through the real bridge, catalog, and schedule service rather than mocked schedule failures.
Co-authored-by: Felipe Fidelix <felipe@fdx.is>
* fix(server): distinguish caller errors from stored data failures
Tag validation failures where tool arguments are parsed, so schema errors from persisted state remain internal failures. Preserve the missing-agent error at its producer and translate it for prompt delivery. Real HTTP regressions fail for both misclassifications before this change.
* fix(server): preserve messages for unclassified tool failures
Keep existing error messages on HTTP 500 responses so agents can correct failed tool calls. Exercise missing workspaces through the real script service and bridge route.
* fix(server): preserve clearing schedule names with blank strings
---------
Co-authored-by: Felipe Fidelix <felipe@fdx.is>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(android): preserve dictation across audio interruptions
Playback focus loss does not mean microphone capture is unavailable. Keep dictation alive across media takeovers and wait for Android input silencing to clear, while preserving conversation interruption protection.
* test(android): verify dictation clears its selected audio route
* fix(app): make home actions reachable on short screens
Keep the home cards and community links in one scrollable flow so overflow stays reachable and the footer cannot overlap the cards.
* fix(app): keep home scrolling clear of window drag regions
Replace the upload-only submission with the existing TestFlight distribution job shape so external testers receive stable releases. Keep App Store review targeting the same version and build without a second binary upload.
A pending thinking change retired the live query when mode, model, or command controls requested it. Reuse the query owned by the current turn and apply pending launch settings after the turn finishes.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(server): run slash commands sent by another agent out of band again
* fix(omp): send /steer and /follow-up as prompts so an idle agent starts a run
* test(omp): verify idle follow-ups and parent notifications
Cover real daemon and MCP delivery, and keep the fake runtime tests independent of an installed OMP executable.
Co-authored-by: Hung Le <22406703+nhthung@users.noreply.github.com>
* fix(omp): report idle follow-up admission failures
Use the existing prompt lifecycle for an idle command so acknowledgements and late rejections settle the child and notify its caller. Keep busy commands on the streaming prompt path, and accept one runtime prompt input object.
---------
Co-authored-by: Hung Le <22406703+nhthung@users.noreply.github.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(claude): show a background subagent's action log on its card
Claude Code answers a background Agent call at launch and streams the
child's frames afterwards. Since #6067 the sidechain tracker stopped
emitting the parent card once the call had its result, so the card's
log stayed empty. The tracker now rebuilds the card from the settled
call, keeping its status, output, and label, and adds the child's
action log.
Closes#6157
* fix(claude): release a settled subagent call when the child finishes
* fix(claude): keep a background subagent's card in the turn that settled it
A background child often keeps streaming after the parent's turn ends. Its
frames then rebuilt the settled card in the next turn with an empty log, and
the projection showed it as a second card for the same call. The settled call
now rebuilds the card only within the turn that answered it.
* test(claude): model the turn boundary in background card coverage
---------
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(app): render terminal-profile icons instead of the generic Bot
* test(app): verify terminal profile artwork in the launch menu
Exercise the rendered SVG, dimensions, theme color, catalog artwork and fallback through the real browser app instead of inspecting hook results in JSDOM. The browser regression fails on main before the lookup fix.
---------
Co-authored-by: Noah <noahguerin20@gmail.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Keep keyboard translation attached to the native animation graph when fullscreen resizes the dock. Apply the enabled mode on the native animation frame so Fabric does not restore the previous transform during layout.
* Split the inbound WebSocket validator per message type
zod-aot compiled WSOutboundMessageSchema into one 11 MB function. Every
inbound frame, heartbeats included, ran through it, so V8 hosts (daemon
clients, plugin hosts, the CLI, Electron) eventually Maglev-compiled the
whole function: about 800 MB of compiler memory and multi-second stalls,
again after each deoptimization.
Generate a tag-only envelope validator plus one validator per session
message, and dispatch on the message type. Issue paths, defaults and
passthrough behave as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Use forward slashes in generated validator imports
path.relative() returns backslash paths on Windows, which break the
import specifiers written into the generated files.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Declare the session export helper as a function
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
An iOS pan waiting for manual activation can deliver finger-up without finalizing, leaving the row active and the outer scroller locked. Fail a pending pan when its touch stream ends so the existing finalization path cancels the drag. Track activation immediately so an active drag keeps its normal completion path.
Based on the touch-release diagnosis in #5586. The native regression fails before the fix and passes on the Namespace iOS simulator.
Co-authored-by: Kassam Khoja <kassam@shopwrk.com>
Generated image Markdown was concatenated into the preceding paragraph and rendered at inline size. Preserve paragraph boundaries around image output while retaining intentional inline image behavior.
Cursor agents never showed plan usage cards under their context details.
ACP sessions had no usageSession(), and the Cursor usage source only ran
global discovery, so per-agent discovery always came back empty.
Cursor ACP sessions now report a usage session (provider "cursor" plus
their launch env). The Cursor usage source resolves the cursor-agent
auth.json for that env. Sessions on CURSOR_API_KEY or CURSOR_AUTH_TOKEN,
the in-memory store, or the macOS Keychain (the default there) report
no login.
Reopening a retained sidebar rebuilt its entry collection and invalidated every project row despite unchanged workspace data. Preserve collection identity as well as row identity, while keeping membership and ordering changes observable.
* fix(app): group native assistant prose into selectable runs
* fix(app): render selectable native prose with enriched markdown
* docs: explain native prose selection boundaries and tradeoffs
* fix(app): preserve markdown semantics and render limits in prose runs
* docs: record native prose compatibility fixes and heading limits
* fix(app): allow additive task lists and admonitions in native prose
* fix(app): preserve inline image formatting and reference scope
* fix(app): preserve markdown source and file actions in native prose
Prepare original source blocks with source-position edits from the assistant parser. Keep reference scope and live/history projection consistent, and retain link text when dispatching workspace file actions.
* fix(app): preserve native prose entities and links while bounding preparation cost
* fix(app): preserve provisional syntax at cached reveal boundaries
* fix(app): bound prose append work and preserve markdown edge cases
* fix(app): preserve home paths and normalized prose links
* fix(app): preserve parser-owned links and streaming tilde tails
* fix(app): keep growing markdown on the streaming block renderer
* fix(app): preserve bracketed IPv6 links in native prose
* fix(app): skip prose classification for blocks over the render cap
* fix(app): share selectable native prose across Markdown surfaces
* fix(app): preserve Markdown containers before grouping prose
* revert(app): remove shared prose-run renderer changes
Reverts ab4737bb2 and ac47e1063 before replacing the Markdown engine across all platforms.
* feat(markdown): use enriched renderer across all platforms
* test: verify library image controls and isolate website config
* fix(markdown): preserve file links and image revisions
* fix(markdown): restore diagrams and preserve link origin
* fix(markdown): retain diagram identity and reference scope
* fix(markdown): avoid empty web history measurements
* fix(markdown): preserve authored empty link titles
* test(markdown): exercise find across whole-message history
* fix(markdown): preserve encoded image paths after rebase
Keep the shared renderer image normalization and file-read regression coverage that originally lived in the main integration merge.
* fix(markdown): derive table and task list colours from the theme
* fix(markdown): restore selection highlight on web
* fix(markdown): restore file link hover tooltips on web
* fix(markdown): normalize inline HTML before rendering
* fix(markdown): create renderer types when applying package patch
The published nightly omits both MarkdownRenderers type files. Represent them as new files so patch-package succeeds on a fresh npm install, matching the already patched development package.
* fix(markdown): keep the renderer tree stable while images resolve
* fix(markdown): size inline images to the line and collapse failed loads
* fix(markdown): normalize only parsed HTML nodes
* fix(markdown): keep the first image reference definition
* fix(markdown): preserve complete assistant messages on native
* fix(markdown): classify web links explicitly and complete renderer types
* fix(markdown): preserve spacing between HTML elements
* fix(markdown): normalize HTML following code elements
* fix(markdown): preserve literal image fallback labels
* fix(markdown): retry transient image decode failures
* docs(markdown): record native selection and image limitations
* fix(markdown): preserve first definitions across Mermaid segments
* fix(markdown): reserve pending images and harden web rendering
* test(markdown): cover state retention during image arrival
* fix(markdown): preserve reader position while responses grow
Keep the same message component boundary when a live row becomes retained history, preserving Mermaid runtimes and selection nodes. Observe paced child text commits at the web viewport so sticky-bottom follows before the next frame.
* test(markdown): require growth while holding a streaming paragraph
* test(markdown): isolate image arrival from reload positioning
Mounted row totals change as history virtualizes, so observe the new response and tool directly. Hold the real PNG response after reload until the reading position settles, then verify its reserved height and viewport instead of attributing initial bottom-follow positioning to image decoding.
* fix(markdown): preserve adjacency around inline HTML
* fix(markdown): keep summary block boundaries outside bold text
* perf(markdown): scan HTML block boundaries incrementally
* test(markdown): gate delivery while establishing a held reader
* test(markdown): name shared renderer setup and file actions
* test(markdown): assert linear HTML boundary work without timing
* test(markdown): check large HTML table output without observing built-ins
* fix(markdown): restore accentBright link colour
* fix(markdown): require letters and known extensions in bare path recognition
* fix(markdown): match main's blockquote, bullet and h6 colours
* fix(android): keep inverted chat still during text selection
React Native decomposes Android list inversion into a 180-degree rotation with positive scaleY, so the existing visibility-scroll guard never recognized the inverted viewport. Inspect the composed matrix and reproduce the actual list transform through the native view manager in regression tests.
* fix(markdown): linked inline code keeps inline code size
* fix(markdown): reserve a placeholder box with an image glyph while images load
* fix(markdown): bare prose paths need an extension, line suffix or depth
Slash-separated English pairs were recognized as file links in prose. Require stronger path intent there while preserving the existing inline-code recognition rule.
* fix(markdown): horizontal blocks own their drags on compact layouts
* perf(markdown): keep native streaming updates in the unfinished tail
Native markdown reparses each updated view in full. Keep parser-closed blocks in a memoized head during paced reveal, then reunite completed prose for cross-paragraph selection. Preserve trailing spacing and source indentation at the split.
* perf(markdown): reuse Android measurements across view mounts
* test(markdown): bare Fabel/model and daemon/client stay plain text when copied
a502cfd0d stopped linking bare two-segment prose paths; the clipboard
expectation still carried the links from the earlier loose pattern and
failed on CI (playwright shard 1/4 on 60b4e2d05). main expects the plain
text, so restore that.
* perf(app): limit native chat mount batches
Whole-message Markdown rows measure synchronously during Fabric layout. Render fewer history rows per batch and retain a smaller offscreen window so opening a chat does not measure twelve history rows at once.
* fix(app): keep populated markdown in sync with native layout
Fabric can collapse streaming blocks before asynchronous rendering replaces the old text, producing a blank frame. Commit populated-view updates synchronously while retaining worker preparation for first mounts to avoid moving large history parsing onto the UI thread.
* perf(app): preserve unchanged sidebar workspace projections
Reopening a retained sidebar rebuilt its entry collection and invalidated every project row despite unchanged workspace data. Preserve collection identity as well as row identity, while keeping membership and ordering changes observable.
* Revert "perf(app): preserve unchanged sidebar workspace projections"
This reverts commit a3919faed4.
* fix(app): show press feedback on sidebar rows, footer icons, and settings tiles
Touch has no hover, so these controls gave no acknowledgement when pressed.
Rows take the workspace row's pressed fill; the sidebar header row paints it
on the row so the trailing slot lights up with the button. Footer icon buttons
take the round header-control highlight.
* style(app): format sidebar header row
* fix(app): keep footer icon corners tappable on web
Web hit-testing follows border-radius, so rounding the footer buttons turned
their 44pt square targets into circles and taps near the corners fell through.
Paint the round highlight as a layer inside the square button instead.
* fix(app): preserve background workspace visibility after reload
Persist the background flag and reject workspace rows that predate it so the existing cache repair clears their workspace cursor and refreshes the baseline.
* test(app): make legacy workspace cache read paths explicit
Header islands lose their filled background and shadow; the round tap
highlight stays. The floating header's fade now carries the icon contrast
the island fill used to give, so it stays solid through the header and
runs a longer tail into the content. The fade lives inside the overlay
because Android clips children that overhang their parent, and keeps
three gradient stops because Android rendered a four-stop version far
more transparent than specified.
Single-line header titles read one step larger. The compact composer
gets a soft drop shadow. The compact settings root groups App and Host
into islands of tiles, driven by breakpoint style values so desktop
keeps its plain list.
Limit implicit pane and remount focus replay to the existing fine-pointer policy. Fullscreen keeps one editor mounted, so touch navigation does not need the restoration introduced for the former portal handoff.
* feat(app): run and watch workspace scripts from the compact header menu
The compact header redesign dropped the mobile scripts button, so phones had
no way to run, stop, or restart workspace scripts. Scripts are now a page of
the ⋯ sheet: the root row shows what is running, tinted by health, and the ⋯
carries a dot while a running service is unhealthy. Desktop keeps its popover;
both surfaces share one set of script actions and one status model.
Script status also froze on native whenever the sidebar was closed and no
agent tab was viewed. DirectorySynchronizer releases its subscriptions,
including script_status_update, when nothing demands the directory, and the
workspace screen never demanded the workspace it shows. Web hid this because
the favicon sync holds demand for the whole session. The focused workspace now
holds directory demand, which also lets Restart's second half (start once the
daemon reports the script stopped) complete.
* perf(app): keep workspace script status live without syncing the directory
The focused workspace held full directory demand so the scripts menu saw
script_status_update on phones with the sidebar closed and no agent tab.
That fetched the host's agent and workspace lists on every focus and
streamed every agent and workspace update for the whole host.
Script status now has its own event subscription, which a workspace
screen can hold alone. It fetches the workspace's scripts once when the
subscription lands, because the subscription only carries changes.
Directory demand also holds it, so the sidebar and agent tabs see the
same updates as before.
Android's scroll guard now reads the view's transform matrix, so React Native's inverted chat list (scale: -1, decomposed into a 180-degree rotation with positive scaleY) no longer scrolls on text focus or selection.
Closes#7125
Co-authored-by: luke <luopeihuan@gmail.com>
Co-authored-by: RoundRedDot <roundreddot@gmail.com>
* fix(app): let Escape close an open popover before it interrupts the agent
The agent.interrupt shortcut listens on the window in the capture phase, so
it sees Escape before anything else. Registered overlays get first refusal
through dispatchTopWebOverlayKeyDown, but plugins cannot register, and draw
their floating panels with the Popover API instead. With such a panel open
and the composer focused while an agent runs, Escape closed the panel and
interrupted the agent in the same press.
While a popover is open, Escape is left to it and no app shortcut runs. A
second Escape, with nothing open, interrupts as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: RoundRedDot <roundreddot@gmail.com>
* test(app): cover Escape with a popover open through the shortcut path
The first test only checked hasOpenDomPopover and would still pass with the
guard removed from useKeyboardShortcuts. The check is now keyBelongsToDomPopover,
called by the hook, and a browser test drives it the way the app does: a
window capture listener for shortcuts, a popover closed by its own document
Escape listener, focus in a textarea. The first Escape closes the popover
with no interrupt; the second interrupts. Both new tests fail with the guard
removed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: RoundRedDot <roundreddot@gmail.com>
* test(app): exercise plugin Escape through the real composer
Keep the contributor fix and replace the substitute shortcut-listener test with an installed-plugin browser regression. The first Escape must close the preview without cancelling the turn; the second must cancel it.
Co-authored-by: luke <luopeihuan@gmail.com>
Co-authored-by: RoundRedDot <roundreddot@gmail.com>
* test(app): check dismissed model pickers leave the viewport
---------
Co-authored-by: luke <luopeihuan@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: RoundRedDot <roundreddot@gmail.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(app): scroll the Add Project results to the keyboard-selected row
The results list had no scroll ref, so ArrowUp/ArrowDown moved the selection off-screen; the autocomplete reveal math is now shared through useRevealActiveItem.
* test(app): verify the directory selected beyond the results viewport
Check scrolling in both directions and across wrapping, then confirm Enter opens the selected directory. The previous regression used an aria-selected attribute that result buttons do not render.
Co-authored-by: superheher <heh@vivaldi.net>
* test(app): verify scroll reveal through browser behavior
Keep pure offset checks and cover keyboard navigation, wrapping, and replacing search results in the real Add Project flow.
* fix(app): measure retained Add Project rows when revealing selection
* fix(app): measure scroll content through a native View reference
---------
Co-authored-by: superheher <heh@vivaldi.net>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Recording a digit for tab or workspace navigation replaced the existing
1–9 wildcard with one literal key. Preserve the group when capturing a
replacement and display its range after saving and reloading.
Adapted the number-group fix from getpaseo/paseo#5237, with independent
failing-first renderer reproduction and tab/workspace matcher coverage.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Co-authored-by: dbunk903 <74584941+dbunk903@users.noreply.github.com>
* perf(app): overlay pending replica rows on read instead of forcing persists
readRows flushed every pending change to storage, read, and retried
whenever the host revision moved. Every stream reducer flush (48 ms) and
every agent delta advanced that revision, so opening another agent or
workspace on a host with a streaming agent re-serialized the whole
pending batch per iteration and could spin until the stream paused.
Reads now run on the write queue, so no drained batch is mid-apply while
a read runs, and rows still pending in memory are overlaid for the
requested keys. A read never forces a persist and never waits for the
host to go quiet.
Persist retries after a storage failure back off exponentially (1 s to
60 s) and log the failure instead of re-serializing every second.
Before: one readAgent during a 200-commit stream cost 201 store reads and
200 forced persists; a permanently failing store was written 30 times in
30 s. After: 1 read, 0 persists; 5 writes in 30 s.
Fixes#4813
Refs #4814
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* fix(app): log replica persist retry failures
---------
Co-authored-by: Ivan Malison <IvanMalison@gmail.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Co-authored-by: paseo-bot[bot] <paseo-bot[bot]@users.noreply.github.com>
* perf(app): stop re-walking the timeline on every streamed delta
Every coalesced delta rebuilds the stream render model with the same
tail. The model re-sliced the tail at the history-window start on every
call, so the rendered tail was a fresh array per delta and every WeakMap
keyed on it (ordering, history split, turn timing, the layout history
cache, row revision) missed each time. Cache the slice by tail identity
and window start.
Turn timing then still walked the whole rendered tail per delta and
allocated a new assistant-timing Map, which the layout memo depends on.
Split the walk: the tail is walked once per tail identity and the head is
walked from the carried open turn per delta. The Map keeps its identity
while the head adds no completed turn.
Adds identity-contract tests and a vitest bench for the per-delta
pipeline. 200 deltas over a 2,000-item timeline, mean per run:
default mounted window 5.68ms -> 2.82ms, history fully revealed
16.95ms -> 1.81ms.
Fixes#4816
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* perf(app): keep the stream view's per-delta work bounded by the head
The chat outline concatenated tail and head, reduced over the result in
the render body, and rebuilt jumpToPrompt on every delta even when the
outline was disabled, which is every native viewport. Memoize the tail's
contribution, scan only the head per delta, do nothing while disabled,
make jumpToPrompt a stable event so the rail's memo holds, and republish
the active prompt only when the index or reading position changes.
The view built a Set of every mounted id plus the live head per delta for
the outline's pending-jump check; key it on mounted history only and
treat live head rows as always mounted. Split plugin projection so a head
change does not re-project the tail, memoize the history-window boundary
lookup, and read the turn presentation's submission flag with some()
instead of allocating a filtered array.
Refs #4816
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* test(app): confirm full history before measuring streaming
---------
Co-authored-by: Ivan Malison <IvanMalison@gmail.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(app): focus chat feed when clicking message content
* test(app): wait for new chat output before checking scroll position
Co-authored-by: Ivan Pantic <panta@talentkit.io>
* test(app): capture the keyboard paging regression
* test(app): verify paged chat text stays in place
---------
Co-authored-by: Ivan Pantic <panta@talentkit.io>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(server): open worktree.terminals when a workspace is created on its own
runWorktreeSetupInBackground only opened the paseo.json worktree.terminals
when its caller passed runAutoTerminals, and only the fork-PR "allow setup"
handler did. workspace.create, the request the New Workspace screen sends,
ran worktree.setup and then skipped the terminals.
Both callers want the terminals, so the flag goes away: terminals follow a
successful setup on a newly created worktree. Reused worktrees stay excluded
by shouldBootstrap, and the agent setup continuation never reaches this
function, so nothing opens twice.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* test(server): run automatic terminal regressions in CI
Include workspace creation in the existing integration command, and append to the terminal marker so repeated command execution fails the regression.
* test(server): make workspace fixtures portable on Windows
Run the setup fixture through a command accepted by PowerShell, await its readiness, and fix the fixture Git line-ending policy. Allow Windows directory locks to settle during cleanup.
* test(server): expose setup failures and await fixture cleanup
Check the public setup status before terminal startup so failures distinguish setup errors from terminal creation. Use asynchronous directory removal so Windows cleanup retries leave the event loop available.
* test(server): start the Windows setup service asynchronously
The Windows regression starts its service but setup stays running, so terminal bootstrap is never reached. Launch the fixture service explicitly with Start-Process and retain the readiness and setup-completion assertions.
---------
Co-authored-by: David Yang <9925+davidyang@users.noreply.github.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(app): preserve usage provider order across refreshes
* test(usage): keep refresh and scrolling journeys stable
Usage reports can follow the long fixture card, so a fixed wheel distance
can scroll past its final window. Target the named windows with wheel
input while preserving the visibility checks. Assert refreshed provider
order through visible names and keep viewport setup in journey helpers.
Co-authored-by: Sleroq <hireme@sleroq.link>
---------
Co-authored-by: Sleroq <hireme@sleroq.link>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
Each streamed text or reasoning chunk was cut from the accumulated
message with slice, which can share the source string's storage. The
timeline store keeps every chunk, so a 128 KB reasoning part retained
about 128 MB: every chunk held the full message as it stood when the
chunk arrived. Long OpenCode v2 reasoning streams exhausted the daemon
worker's heap.
The timeline now emits an independent copy of each suffix, so retained
memory follows the length of the text.
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>
* fix(app): refine provider catalog actions and error feedback
Use compact row tiles and the shared error toast so failed Add actions remain visible while browsing the catalog. Point Hoplite installation instructions at the CLI install section.
* test(app): supply toast context in provider settings tests
Import the toast API independently of its renderer so the provider section only depends on the context it consumes.
Preserve the bundled Hoplite ACP adapter and provisioning transformer, activate it only after Add, and provide a searchable provider grid. Restore the contributor's original icon without the copyright mark.
Co-authored-by: Odin Glynn-Martin <odin.glynn@icloud.com>
Protocol contracts are available through the SDK peer dependency during development, but clean Git installs lack that dependency tree. Accept erased protocol type imports while retaining runtime dependency resolution and SDK ownership checks.
