Installed plugins rendered duplicate UI slot widgets (e.g. the chat-input-actions icon twice) because the append-only slot registry could run a plugin's initialize() again while its prior registrations were still live.
- Make loadPlugin idempotent: revoke a plugin's prior registrations before re-running initialize().
- Fence overlapping/concurrent loads by a per-plugin generation so the newest-initiated load wins and a superseded (older) load can't revoke the successor or append a stale registration.
- Wrap the scoped registry reflectively so every register* method (incl. registerKeybinding) is fenced, never dropped.
Tests: registry-count regression, real-DOM PluginSlot render reproduction, two overlapping-load regressions, and a registerKeybinding-forwarding regression.
* feat: add opt-in plugin auto-updater
Automatically update installed plugins in the background, off by default
and opt-in. Builds on the existing verified install pipeline — an
auto-update is just an automatic reinstall of a newer catalog package_url.
Two-tier toggle on Settings > Plugins: an instance-wide default (SQLite
plugin_settings) plus a per-plugin override (tri-state auto_update on the
plugin record, carried forward on upgrade). The AutoUpdatePoller
(~90-min cadence + a boot sweep) upgrades only plugins that are currently
active, opted in, and report update_available; disabled/errored plugins
stay on their installed version.
New endpoints: GET/PUT /api/plugins/settings and
PUT /api/plugins/:id/auto-update.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(plugins): recheck auto-update eligibility at install and fix test mock
Address PR #1903 review feedback:
- Re-check each plugin's current status + effective opt-in immediately before
InstallFromURL (eligibleForAutoUpdate), closing the TOCTOU window between the
candidate snapshot / catalog fetch and the install so a plugin disabled or
opted out mid-sweep is never reinstalled and reactivated.
- Correct the marketplace spec to describe the real failure behavior (shared
with manual updates) instead of promising a version rollback the install
pipeline does not perform.
- Fix the frontend CI failure: the plugins page test mocked the whole
plugins-api module without the new settings/auto-update functions, so
useAutoUpdateSettings threw at mount.
- Add an aria-label to the per-plugin auto-update Switch and Reset control.
- Add a use-auto-update-settings hook test (load, failure, optimistic setDefault,
rollback) and an eligibleForAutoUpdate unit test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(deps): patch critical Dependabot advisories
Resolve all open critical-severity Dependabot alerts:
- golang.org/x/crypto 0.51.0 -> 0.52.0 (CVE-2026-39830/39831/39832/
39833/39834/42508/46595, GHSA ssh agent/key-constraint bypass, DoS)
- vitest ^1.6.0 -> ^4.1.10 in apps/cli (CVE-2026-47429; aligns with
apps/web, dedupes the lockfile to a single vitest 4.1.10)
- tar ^7.5.11 -> ^7.5.19 in apps/cli (CVE-2026-59873 parse DoS;
resolves to 7.5.21)
Backend builds and vets clean; all 280 apps/cli tests pass on vitest 4.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(cli): bump esbuild to a Vite-8-supported range
The vitest 4 bump pulls vite@8, whose esbuild peer is ^0.27||^0.28,
but the CLI still pinned esbuild ^0.24.0 — pnpm resolved an
unsupported vite/esbuild pair (esbuild@0.24.2) for the CLI test
stack. Bump to ^0.27.0 so vite@8 resolves against esbuild@0.27.2 and
the peer warning clears. CLI tests still pass (280/280).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(plugins): add keybindings and modal window capabilities
Add two independent plugin UI capabilities.
Keybindings: plugins declare keybindings in the manifest under
ui.keybindings ({id, default, description}), validated backend-side
(combo grammar + unique ids, requires a UI bundle) and exposed to the
frontend. Plugin JS binds a handler via registry.registerKeybinding(id,
handler); a global dispatcher fires it. Combos are user-overridable in
Settings > Keyboard Shortcuts (namespaced plugin:{pluginId}:{id}) with
conflict detection against core and other plugins.
Modals: host.openModal({title?, content, size?, dismissible?}) returns a
{ close() } handle and renders in a single host-owned Dialog isolated by
PluginErrorBoundary, cleaned up on plugin unload.
Fixture plugin, PLUGIN-API.md / spec, and a Playwright e2e demonstrate a
keybinding handler opening a modal.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(plugins): address review feedback on keybinding dispatch and validation
Core shortcuts now take precedence over plugin keybindings on a colliding
combo (defaultPrevented cooperation between the two capture-phase
dispatchers), a throwing plugin handler no longer aborts dispatch to other
plugins, and the keybinding combo grammar is tightened: the backend validator
and frontend parser both drop the undocumented control/super modifier
aliases and reject shift+digit/symbol combos that can never actually fire
(KeyboardEvent.key reports the shifted character, not the token).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(plugins): stop conflicting core shortcut from double-firing with plugin keybindings
The earlier defaultPrevented guard only covered the central useAppShortcuts
dispatcher. Per-component core shortcuts registered via useKeyboardShortcut
run in the bubble phase, after the capture-phase plugin keybinding dispatcher,
so a plugin-handled-and-prevented combo still triggered the core action too.
Guard useKeyboardShortcut (and use-editor-keybinds, another capture-phase core
listener) on event.defaultPrevented so exactly one action fires per combo.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(plugins): make core shortcuts always win over plugin keybindings
Panel-search, task-switcher, and other core listeners never checked
event.defaultPrevented, so a plugin bound to the same combo (e.g. Cmd/Ctrl+F)
could still fire alongside them. Instead of patching every core listener,
usePluginShortcuts now computes the set of effective core shortcut combos
(registry, resolved through user overrides, plus FIND_IN_PANEL/SAVE) and
skips invoking any plugin handler whose combo is in that set, so the
matching core listener always handles the event exactly once.
Also adds a hook-level regression test for useEditorKeybinds' existing
defaultPrevented guard, and fixes a leaked capture-phase window listener in
use-keyboard-shortcut.test.ts that could bleed into later tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(web): keep Cmd+K closing the command panel after shortcut-precedence guard
cmdk's built-in vim bindings intercept Ctrl/Cmd+K and +P as list-navigation
shortcuts and call preventDefault() before the event reaches our window-level
useKeyboardShortcut listeners. Once that hook started bailing on
event.defaultPrevented (added to fix core/plugin shortcut double-firing), the
palette's own Cmd+K close/reopen toggle got silently suppressed whenever focus
was inside the open panel. Disable cmdk's vim bindings on the command palette
so its Ctrl/Cmd+K and +P always reach the app-level toggle.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(plugins): document keybindings and modal window plugin APIs
Fill in the public author-facing docs that the internal plugin-keybindings
and modal-window contract (PLUGIN-API.md, spec.md) already covers: the
registry.registerKeybinding / host.openModal JS surface, the ui.keybindings
manifest field with its combo grammar, and a capability mention on the
Plugins overview page.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Subagent cards keyed their auto-expand on `isActive || Boolean(resultText)`.
Silent (Auggie-style) subagents return a result_text but stream no child tool
calls, so `resultText` stays populated forever and the card rendered
permanently expanded, unlike subagents that emit child tool calls (which
collapse to their header + metadata row on completion). A render-phase reset
also wiped any manual collapse the moment result_text arrived.
Key auto-expand on the running state alone and drop the reset, so completed
subagents collapse consistently and the result text is one click away. Add
component coverage for the silent-subagent collapse, expand-on-click, and
manual-collapse-persists paths.
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat: resolve @prompt_name saved-prompt references in workflow step prompts
Workflow step prompts, the GitHub review-watch task prompt, and the CI
auto-fix prompt now resolve "@name" references to saved custom prompts
(Settings > Prompts) at invocation time, appending the referenced
content as hidden context while keeping the mention visible - matching
how chat messages and message_task already resolve @mentions. Editing
a saved prompt now propagates to every workflow step that references
it without needing to update each step individually.
Adds a Monaco "@" completion provider to the step-prompt and
review-watch editors, alongside the existing "{{placeholder}}"
provider.
* fix: address PR review findings on prompt-reference expansion
- Make AppendReferenceExpansions idempotent: a prompt that already
contains a previously-appended expansion block is returned unchanged,
since the block's own "### @name" headers would otherwise be
re-scanned as fresh mentions on a second call.
- Skip "@name" expansion entirely for CLI-passthrough sessions: their
prompts are written raw to a PTY with no <kandev-system> stripping,
so the hidden block would otherwise leak into the terminal verbatim.
* fix: fail safe on passthrough lookup error in startTask
A session-lookup error while resolving isPassthrough for the Kandev
MCP wrap and prompt-reference expansion previously defaulted to
false (non-passthrough), risking a hidden <kandev-system> block
leaking into a real passthrough session's PTY on a transient error.
Extracted resolveIsPassthroughForLaunch, which now fails safe by
defaulting to true (skip injection) and logs a warning.
* fix: scope expansion idempotency marker to system blocks
The idempotency guard added in a prior commit checked for the
expansion marker anywhere in the prompt, which could falsely
suppress a legitimate first-time expansion if caller-controlled
text happened to contain that literal phrase. Add
sysprompt.HasSystemMarker, mirroring the existing HasKandevContext
pattern, so the marker is only recognized inside a formatter-
produced <kandev-system> block. Also strengthens a passthrough test
assertion that didn't actually distinguish the snapshot path from
the fail-safe default.
* fix: match exact expansion-block prefix in idempotency guard
The prior scoping fix accepted the marker anywhere inside any
<kandev-system> block, which could still false-positive against an
unrelated system block (e.g. a plan-mode prefix or MCP context
wrap) that happens to mention the marker phrase without it being
the block's own content. Check for the exact literal prefix this
package's formatter produces instead: the opening tag immediately
followed by the marker, with nothing in between.
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
* fix: follow session switch when moving between different-agent steps
Moving a task to a workflow step with a different agent profile makes the
backend complete the current session and promote a new primary session for
the new agent. The chat UI was supposed to follow that swap, but the
follow-focus in handleTaskUpdated was suppressed whenever the user had
pinned the session they were on (clicking into a session pins it). The pin
was on the very session being retired, so the user stayed stranded on the
now-completed session — still showing the old agent and plan mode.
The activeSessionId === previousPrimary guard already scopes the follow to
"user is sitting on the session being replaced", so the pin should only
protect a *different* still-live session (active-session drift), never the
one being swapped away. Extract shouldFollowPrimarySwap encoding that rule.
Because it keys off task.updated's primary_session_id it is independent of
WS event ordering and does not depend on the new session row having loaded.
* fix: address review feedback on session-switch follow-focus
Scope the follow-focus fix to genuine session retirement and fix the real
root-cause race, per Greptile/Codex/CodeRabbit review:
- Revert the task.updated pin bypass. Manual "Set as Primary" swaps the
primary without retiring the old session, so bypassing the pin there would
yank a pinned user off a still-live session. Restore pin preservation and
add a guard so a first-primary assignment (null previous primary) never
steals focus.
- Fix the actual bug in the session state-transition handoff: when a retired
session has no non-terminal successor in the per-task list yet (the switch
promotes the new primary via task.updated before the old session's terminal
state_changed lands), fall back to the task's authoritative primary. This
path only runs when the active session truly reaches a terminal state, so it
never fires for manual set-primary.
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
* fix(executor): re-clone provider repos when local path is missing or not a git repo
Previously the re-clone guard only fired when local_path was empty, so a
set-but-invalid path (e.g. deleted checkout, or relocated by the provider
clone-path/identity change) was fed straight to the worktree preparer,
which hard-failed with "repository is not a git repository" instead of
recovering.
Two-part fix:
- resolveTaskRepoInfo now also re-clones when local_path exists but isn't
a valid git checkout (new isLocalGitRepo helper), self-healing on next
launch.
- validateWorktreeRequest now fails the "Validate repository" step with an
honest error naming the offending path when it truly isn't a git repo,
instead of passing and cascading into a misleading "sync interrupted"
failure downstream.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(executor): gate stale-path reclone to provider repos and cover resume + multi-repo
Restrict the stale-local-path re-clone to genuinely provider-backed repos
(SourceType != "local") so a local checkout that's temporarily missing is
never silently overwritten. Centralize the re-clone guard into a shared
ensureRepoLocalPath helper and call it from the single-repo resume path
(applyResumeRepoConfig) so resume self-heals like a fresh launch. Apply the
same non-git-path validation used by the single-repo prepare path to the
multi-repo prepareOneRepo path so a stale secondary repo fails fast with a
clear message instead of surfacing worktree.Manager's ErrRepoNotGit.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(web): add app status bar
* feat(web): add app status bar visibility toggle
* feat(web): persist status bar item order
* fix(web): refine app status bar behavior
* fix(web): polish app status bar presentation
* test(web): split plugin slot coverage groups
* fix(web): address status surface review feedback
* fix(web): stabilize status bar overlay tests
* fix(plugins): unload previous version before reloading on plugin update
The install/update path reloaded a plugin without first unloading the
previous version's registrations, unlike disable/uninstall. Because the
slot registry only appends, this ran initialize() twice and left two
main-top-bar slot entries, duplicating the plugin's widget icon until
a page refresh.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(plugins): cache-bust plugin bundle import so updates re-execute
unloadPlugin(id, {evictCache: true}) drops the cached bundle registration
on install/update, but a same-tab re-import() of the identical bundle URL
returns the browser's already-evaluated ES module without re-running its
top-level registerKandevPlugin() call — leaving an updated plugin active
but missing from the UI. Key the bundle URL on the plugin's installed
version (?v=<version>) on both the frontend (toActivePlugin) and the
backend boot payload (bootActivePlugins) so a version bump forces a real
re-import while an unchanged version stays cache-friendly across boots.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(plugins): update bundleUrl assertions for cache-bust query param
toActivePlugin now appends a ?v=<version> cache-busting query param to
bundleUrl; these two assertions still expected the bare path.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(plugins): only evict bundle cache on install/update, not on enable
loadIfActive unconditionally evicted the cached bundle registration, which
broke the disable->enable cycle: since bundleUrl doesn't change on plain
enable, the browser's ES module cache returned the already-evaluated module
without re-running registerKandevPlugin, leaving the plugin's slots/routes
unregistered until a full page refresh.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>