Resolve image attachment paths with realpath before reading so symlinks inside allowed roots cannot escape to arbitrary files. Preserve support for symlinked project/upload roots and skip invalid descriptors instead of failing the provider turn.
Resolve image attachment paths with realpath before reading so symlinks
inside allowed roots cannot escape to arbitrary files. Preserve support for
symlinked project/upload roots and add focused coverage for both cases.
Replace the path.relative guard in isPathInsideDirectory with the
resolve + startsWith(root + sep) idiom so the path-injection barrier
is visible to code scanning; behavior is unchanged.
- enforce a second image trust boundary in the provider builders:
buildClaudeUserContent/buildCodexInputItems only accept files inside
the upload store or the run's working directory (CodeQL path
injection)
- drop an always-true selectedProject check in handleSubmit
- remove the unused resume option from spawnCursor
- validate chat image attachments server-side: only files inside the
~/.cloudcli/assets upload store may reach provider file reads
- harden asset serving with nosniff and attachment disposition for
SVGs to prevent stored XSS
- show the timestamp on image-only user messages
- serialize git stage/unstage calls and defer the status re-sync so
rapid toggles can't interleave or flicker
- use a literal hex fallback for commit ref badges (alpha suffix on a
var() string produced invalid CSS)
- stop binding a URL session to a guening
project instead
- add the missing attentionRequiredIndicator key to all sidebar locales
- clean up dangling conjunctions left the
de/ru/tr/ja/zh-CN/zh-TW READMEs
- render a VSCode-style commit graph in the source control history:
lane-assignment algorithm, SVG strip with colored rails, merge and
branch curves, commit dots, and branch/tag badges per commit
- /commits returns parents and ref decorations across all branches
(--branches --remotes --tags --topo-order) using unit-separator
fields so pipes in commit subjects can't break parsing
- collect stats via a single --shortstat pass instead of one
`git show --stat` call per commit; raise the history limit to 50
- replace child_process spawn with cross-spawn in all runtimes,
routes, and services: it resolves .cmd shims and PATHEXT on
Windows (fixing taskmaster's npx invocations) and delegates to
native spawn elsewhere, removing the per-file win32 ternaries
- unit tests for the log parser and lane assignment
- make staging real in the git panel: new /stage and /unstage endpoints,
/status now reports the actual index via porcelain -z (handles renames,
conflicts, and paths with spaces), and Stage/Unstage All run real git
commands instead of toggling client-side state
- add branch search to the header dropdown and Branches tab
- persist the chosen permission mode per provider so a brand-new chat
keeps it once the session id arrives, instead of reverting to default
- replace cmdk's fuzzy model search with strict token matching so
"chatgpt" no longer surfaces unrelated models
- unit tests for the git status parser
* feat(voice): add optional speech-to-text input and read-aloud TTS
Adds a push-to-talk mic button in the composer and a read-aloud button on
assistant messages. Both are opt-in and hidden unless a voice backend is
configured via VOICE_SIDECAR_URL.
The auth-gated /api/voice proxy forwards to a configurable backend exposing
/transcribe and /tts (provider-agnostic); the frontend probes /api/voice/health
and hides the controls when disabled. Adds i18n keys and docs/voice.md.
Includes a local, no-API-key reference backend in voice-sidecar/ (faster-whisper
for STT, Kokoro-82M for TTS, both CPU-capable).
* refactor(voice): provider-agnostic backend and in-app config
Switches the voice proxy to the OpenAI audio API (/v1/audio/transcriptions and
/v1/audio/speech) so it works with OpenAI, Groq, or a local server. Adds a
Settings -> Voice tab (base URL, API key, models, voice) plus a Quick Settings
toggle, and removes the bundled Python sidecar.
Review fixes: stop mic tracks on unmount, clear the global TTS stop handler and
revoke leaked blob URLs, add fetch timeouts in the proxy, surface mic errors in
the button, trim before appending transcripts, and drop the repo-wide wav ignore.
* fix(voice): relax backend timeout and surface timeout errors
Bumps the proxy timeout to 5 minutes (VOICE_TIMEOUT_MS) since local TTS can
synthesize long messages at roughly real-time, and returns a clear timed-out
message (504) instead of failing silently. The read-aloud button now shows
backend errors.
* fix(voice): play read-aloud through an app-level player to stop cutoffs
Read-aloud now runs in a single module-level player outside the React tree instead
of per-message component state. Switching chats or re-rendering a message no longer
revokes the blob URL mid-play (the 'Invalid URI' cutoff). Adds content-keyed caching so
re-listening doesn't regenerate, and reuses one audio element (also unlocks iOS once).
* fix(voice): address review (SSRF guard, auth mapping, client timeout)
Validates the user-supplied backend URL (http/https only, blocks the link-local
metadata range) to prevent SSRF; remaps upstream 401/403 so a bad voice API key
isn't read as the app's own auth failing; adds a client-side AbortController timeout
on the read-aloud request so the button can't sit in loading if a request stalls.
* docs(voice): provider-agnostic wording and jsdoc on proxy functions
drop leftover sidecar/faster-whisper references now that the backend is any
openai-compatible voice api, and add jsdoc to the voice-proxy functions so the
docstring coverage check passes.
* fix(voice): harden timeout parsing, tts input check, and player abort
- fall back to the default when VOICE_TIMEOUT_MS is non-numeric or <= 0, so a
bad override can't make the abort fire immediately
- type-check the tts `text` before calling .trim() so a non-string body returns
400 instead of throwing
- abort the in-flight TTS fetch on stop() and on a superseding play, so tapping
read-aloud repeatedly doesn't leave orphaned requests generating audio
* feat(voice): send transcript with the main send button while recording
while dictating, the main send button stops recording, transcribes, and sends
in one tap, matching the codex-style flow. the mic button still stops and drops
the transcript into the input box to edit before sending. voice recording state
is lifted into the composer so both buttons share it, and the send button is
enabled (not grayed) while recording. also fix a pre-existing type error: the
quick-settings preferences map was missing voiceEnabled.
* fix(voice): make stop() idempotent so a double tap can't throw
guard on the recorder's own state instead of react state, so a double tap or
the mic and send buttons both firing won't call stop() on an already-inactive
MediaRecorder.
* fix(voice): expose TTS format in user settings
* fix(voice): harden recording and backend behavior
Redirects could bypass the backend URL guard, and TTS playback waited for full buffering.
Recording could overlap or finish after teardown. Controls also ignored backend readiness.
Explicit formats and config-aware cache keys prevent stale audio after settings change.
* fix(voice): validate config and request boundaries
Malformed stored settings could break voice requests instead of using safe defaults.
Health results could outlive auth changes. URL checks also did not guard the fetch sink.
Remove constant recorder branches so lifecycle cancellation stays clear.
* fix(voice): separate client and server backends
User-selected backend URLs must remain usable without letting clients control server requests.
Call custom providers from the browser while keeping the server proxy bound to its configured host.
This restores voice controls for frontend settings without reopening the SSRF path.
* fix: hide voice options until enabled
---------
Co-authored-by: newsbubbles <nathaniel.gibson@gmail.com>
Co-authored-by: Simos Mikelatos <simosmik@gmail.com>
Interactive shells could resolve bundled or system CLIs before user-installed npm binaries.
Move existing user npm global directories to the front of PATH while preserving all other entries.
Co-authored-by: Simos Mikelatos <simosmik@gmail.com>
When the package is updated on disk but the long-lived server process is
not restarted, the new frontend bundle (served from disk) talks to the
old running backend. New DB-backed features then fail silently — e.g.
deleting/archiving a session appears to do nothing — because the new
schema/routes only take effect on restart.
Nothing currently detects this skew: useVersionCheck only compares the
frontend's build-time version against the latest GitHub release.
This exposes the running server's version (captured once at startup) via
/health, compares it to the frontend's build-time version in
useVersionCheck, and shows a "restart required" banner in the sidebar
(and a small indicator in the collapsed sidebar) when they differ.
- server: add `version` (RUNNING_VERSION, read once at startup) to /health
- useVersionCheck: return `restartRequired` / `runningVersion`
- SidebarFooter / SidebarCollapsed: surface a restart-required banner
- i18n: add `version.restartRequired` to all 10 sidebar locales
Verified with `tsc --noEmit` (client + server) and eslint.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Simos Mikelatos <simosmik@gmail.com>
* feat(skills): add provider skill management
Users need one settings surface to discover and install skills without manually navigating provider-specific directories.
Add provider-backed global skill installation for Claude, Codex, Gemini, and Cursor, while keeping OpenCode read-only because it reuses other providers' skill locations.
Add a responsive Skills settings tab with scoped discovery, search, refresh controls, markdown and folder uploads, upload feedback, and overflow-safe layouts.
Validate bundled skill files and paths before writing them, preserve scripts and assets, and cover provider discovery and installation behavior with tests.
* fix(skills): preserve uploaded skill folders
Folder drops discarded supporting scripts and assets.
Keep relative paths and upload every file from the selected skill folder.
Use the selected folder name for installation and cover it in provider tests.
* fix(skills): restrict standalone skill uploads
Only show Markdown files when selecting standalone skills.
Normalize browser file paths so SKILL.md is not mistaken for a folder named dot.
* fix(skills): validate installs before writing
Preserve bundled files and normalize fallback names across skill installation paths.
Validate complete batches before writing and reject existing targets to avoid partial installs.
Keep project metadata and make folder selection tolerant of casing and cancelled dialogs.
* fix(skills): overwrite existing installations
Replace an existing skill directory instead of rejecting a duplicate installation.
Remove stale supporting files so the installed directory exactly matches the new upload.
The session indexer scans ~/.claude/projects recursively via
findFilesRecursivelyCreatedAfter, which descends into per-session
subagents/ directories. Claude writes subagent transcripts at:
~/.claude/projects/<encoded-cwd>/<session-id>/subagents/agent-<id>.jsonl
These files repeat the parent session's sessionId. When indexed as
standalone sessions they upsert over the parent row and overwrite its
jsonl_path with the subagent path, corrupting the main session record
(the sidebar then points at, and renders, the subagent transcript).
Add a single isSubagentTranscript() guard (path segment named
"subagents") and apply it in both the recursive scan and the
single-file watcher path.
Co-authored-by: Haile <118998054+blackmammoth@users.noreply.github.com>
Session history and token usage reads already have a stable app session id.
Passing provider and project hints from the frontend kept those reads coupled
with provider-specific state that the backend can resolve from the session row.
Resolve token usage provider server-side and narrow the session store read API
to session id plus pagination. This keeps provider-specific storage decisions
behind the backend boundary and makes reconnect, pagination, and load-all use
the same session-owned contract.
The sidebar had to understand cursorSessions, codexSessions,
and other provider buckets because /api/projects exposed
provider-shaped arrays.
That leaked backend adapter storage into project state and made
frontend behavior drift each time a provider needed another bucket
or exception.
Return one sessions list with provider metadata instead. Project
state, search, and running-session filtering now share one contract,
while provider-specific storage remains behind the backend boundary.
The remote environment could start OpenCode runs under /opt/claudecodeui.
That happened even when the selected project path was correct.
The integration relied on child-process cwd alone.
OpenCode run resolves its workspace through the explicit --dir contract.
Pass --dir with the resolved working directory.
Assert in the CLI test that launch args include the workspace dir.
The sidebar could keep a provider-native id after backend remapping.
That left a duplicate non-working session visible until refresh.
Fresh sessions could also appear hours old.
SQLite CURRENT_TIMESTAMP is UTC without a timezone suffix.
Browser parsing then treated those values like local time.
Broadcast a canonical session_upserted event when the provider id is mapped.
Collapse provider-id aliases onto the stable app session id in the client.
Normalize session-row timestamps to ISO UTC when reading from the repository.
Add a running-session view to the sidebar, including header controls, running counts, empty states, and row-level processing indicators so active provider work is visible outside the current chat.
Hydrate running state after refresh through a status-only /api/providers/sessions/running endpoint backed by chatRunRegistry.listRunningRuns, then sync and poll the frontend processingSessions map from AppContent without attaching to chat streams or replaying messages.
Preserve fresh local processing entries during sync so newly sent messages are not cleared before the backend registry catches up, and clear completed sessions once the status endpoint no longer reports them.
Thread active session state through sidebar project/session components, show rotating loaders for processing sessions, and keep the running search mode expanded and filterable.
Fix optimistic local user-message dedupe so repeated prompts are only collapsed when a matching server echo appears from the same send window, preventing sent messages from disappearing until assistant completion.
Add registry test coverage for listing currently running app sessions.
Tests: npx eslint on changed files; npx tsc --noEmit -p tsconfig.json; npx tsc --noEmit -p server/tsconfig.json; npx tsx --tsconfig server/tsconfig.json --test server/modules/websocket/tests/chat-run-registry.test.ts.
The frontend previously juggled placeholder IDs, provider-native IDs, and session_created handoffs, which caused race conditions and provider-specific branching. This introduces app-allocated session IDs, a chat run registry with event replay, delta sidebar updates, and one kind-based websocket contract so the UI can treat every provider the same while JSONL remains the source of truth.
Replace the chat processing banner with a minimal activity indicator and
rebuild the state model underneath it. The old banner was driven by five
overlapping pieces of state (isLoading, canAbortSession, claudeStatus in the
chat, plus two app-level Sets updated in lockstep through four callbacks)
that had to be kept in sync imperatively. Because completion and status
events mutated the *viewed* session's flags regardless of which session they
belonged to, a background session finishing could hide the indicator for a
still-running session, returning to a finished session could briefly show a
stale banner, and a late status reply could override a newer request.
The fix is structural rather than patch-by-patch: a single
Map<sessionId, {statusText, canInterrupt, startedAt}> in useSessionProtection
is now the only source of truth for "this session is working". The indicator,
stop button, composer streaming state, and session protection are all derived
from the viewed session's entry on render, so there is no stale local copy to
restore or reset when switching sessions. A PENDING_SESSION_ID sentinel
covers the window before a new conversation receives its real session id.
Terminal events delete the entry atomically, which is why the indicator
disappears the instant the final chunk arrives. Stale check-session-status
replies are discarded via an ifStartedBefore guard (an idle reply older than
the entry's startedAt describes a previous request, not the current one).
The second half unifies the provider lifecycle contract, because the frontend
could not be made race-free while each provider terminated differently:
- cursor emitted complete twice per run (result line + process close), which
double-played the completion sound and let a late close-complete clear a
newer request's indicator
- aborts produced two completes (the abort-session reply plus the provider's
own non-aborted one), so cancelling a run played the celebration sound
- codex omitted exitCode; others attached ad-hoc fields (resultText, isError,
isNewSession) the client had to know about
- claude/codex failures ended with only an error event while gemini/cursor
also emit kind:'error' for mid-run stderr noise, so 'error' was ambiguous
between "the run died" and "a process wrote to stderr"
Every run now ends with exactly one complete built by createCompleteMessage()
({sessionId, actualSessionId, exitCode, success, aborted}); abort-session
sends it on behalf of cancelled runs and providers detect the abort and skip
their own. error is demoted to an informational row, so stderr noise no
longer kills the indicator mid-run, and the client celebrates only
success: true completes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Users can miss chat completions while the app is in the background.
They can also miss completions when their attention is elsewhere.
Add opt-out sound notifications and a temporary title marker.
This makes completion noticeable without external audio assets or persistent browser notifications.
Plugin servers are started with a deliberately minimal env (PATH, HOME,
NODE_ENV, PLUGIN_NAME). On Windows that drops system variables that child
processes need to bootstrap. The one that bit me: without APPDATA, CPython
cannot find the per-user site-packages, so a plugin that shells out to a
pip install --user CLI launches the tool but it dies with ModuleNotFoundError.
SystemRoot, PATHEXT and TEMP cause similar failures for other tools.
On win32, pass through a small allowlist of non-secret system variables
(SystemRoot, windir, SystemDrive, USERPROFILE, APPDATA, LOCALAPPDATA, TEMP,
TMP, PATHEXT) when they are set. No change off Windows, and no host secrets
are exposed.
Replace bare background operator with nohup+disown so the cloudcli
server process survives after the sbx exec session terminates.
Also redirects stdout/stderr to /tmp/cloudcli-ui.log for debugging
via `cloudcli sandbox logs`.
Fixes#791
Co-authored-by: NoahHahm <noah@naverz-corp.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Haile <118998054+blackmammoth@users.noreply.github.com>
Users need a visible upload path from the explorer itself, not only drag and
drop behavior with no progress feedback. Routing picker and drop uploads
through one XHR-backed hook keeps progress, validation, refresh, and success
counts consistent for every upload source.
The 200MB limit is mirrored in the client, multer, and nginx template so large
uploads fail predictably instead of being blocked by whichever layer sees the
request first. The server also returns explicit requested and uploaded counts
so partial or multi-file batches can render accurate status text.
