* feat: add voice mode for chat input
Adds a voice input button to the chat composer with three transcription engines:
in-browser Web Speech API, in-browser Whisper via @huggingface/transformers, and
a server-side OpenAI Whisper fallback gated on KANDEV_VOICE_OPENAI_API_KEY.
User-configurable from a new /settings/voice-mode page:
- Master on/off toggle (defaults to on; opt-out, not opt-in)
- Engine preference (auto / Web Speech / Whisper Web / Whisper Server)
- Language hint (auto-detect or BCP-47 pinned)
- Activation mode (click-to-toggle or hold-to-talk)
- Auto-send after transcription
- Whisper Web model size (tiny / base / small)
- Editable Cmd+Shift+M keyboard shortcut
Voice settings live under user_settings.voice_mode as a nested JSON object —
no schema migration. Setting changes broadcast via the existing
user.settings.updated WS event so any open chat tab updates live.
The mic button is rendered on the right of the composer alongside the send
button (round primary fill; destructive fill with pulse ring while recording).
On browsers without a usable engine (e.g. iOS Chrome over HTTP) the button
stays visible as a muted greyed icon — tapping it surfaces a toast with the
reason (typically a missing secure context).
Whisper Web runs in a Web Worker with fp16 decoder weights (avoids the
MatMulNBits scale-missing crash that q4/q8 hit on WASM). Models are pulled
from the onnx-community/* mirror and cached by transformers.js in the Cache
Storage on first use.
* fix: address voice mode PR review feedback and E2E selector clash
Bot-review fixes from #1159:
- Map http.MaxBytesError from c.FormFile to 413 instead of bare 400.
- Strip BCP-47 region suffix before passing language hint to Whisper Web.
- Recover from worker crash: terminate + null worker + reset ready flag so
next init/transcribe creates a fresh worker instead of hanging.
- Read live store state via storeApi.getState() in voice-mode settings
savers to avoid stale-closure clobbering concurrent settings updates.
- Detach Web Speech callbacks before abort() so trailing onend/onerror
events don't mutate hook state after cancel().
- Spread caller init before method/body in voice-api fetch so a caller-
supplied init can't override the multipart POST.
- Replace t.Fatalf with t.Errorf+return inside test HTTP handler goroutine
(FailNow from non-test goroutine is unsafe).
- Add Enabled/AutoSend zeroing assertions in partial-update test so the
bool-overwrite-on-PATCH behavior is locked in explicitly.
E2E fix:
- toolbar-overflow spec used button.rounded-full which now matches both
the send and voice buttons (strict-mode violation). Switched to the
submit-message-button testid for an unambiguous locator.
* fix: scope voice settings rollback + whisper worker error to changed keys
Round 2 of PR review feedback on #1159:
- voice-mode settings rollback now restores only the keys patched by the
failing request (not the whole snapshot), preserving concurrent edits
to unrelated voiceMode fields.
- Shortcut rollback diffs previous vs next and reverts only the changed
keys (restore prior value or delete if absent), preserving other
concurrent shortcut edits.
- Whisper worker error handler captures the worker reference at attach
time and only clears refs when this.worker still matches it — a stale
error from a previously replaced worker no longer nukes the active one.
- Update stale whisper-web-models JSDoc that still pointed at Xenova
mirrors.
* fix: voice mode progress scale, hold-mode race, race-aware rollback
Round 3 of PR review feedback on #1159:
- Normalise Whisper progress to 0–1 in ensureWhisperClient (transformers.js
emits 0–100, the button's display + the `ready: 1` convention expect
0–1 — fixes the 5000% mid-download display).
- Claim driverRef synchronously at the top of finishCapture so concurrent
pointerup + pointerleave invocations early-return instead of racing and
clobbering a freshly-started recording's ref.
- Race-aware rollback: voice-mode and shortcut catch branches now skip
reverting a key when latest[key] !== next[key]. A failed earlier request
no longer overwrites a later successful save to the same key.
- Update the Worker wire-protocol comment to show onnx-community/whisper-base
instead of the stale Xenova/* example.
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
* refactor(worktree): drop unused flat-layout worktrees/ path
Every worktree creation now routes through the task-dir layout
(~/.kandev/tasks/{title}_{suffix}/{repo}/); the legacy flat layout
under ~/.kandev/worktrees/ has been unreachable since the subtasks
PR populated TaskDirName/RepoName unconditionally. Removes the dead
Config.BasePath, Manager.createWorktree, and the flat reconciler.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(worktree): remove unused Reconcile, add guard regression test
Address PR review feedback:
- Reconcile (and reconcileTasksDir / isOrphanedTaskDir) had no callers
anywhere in the tree after the flat-layout removal. Deleted rather than
left dangling; if startup orphan cleanup is wired up later it can return
alongside its actual call site. Also moots the stale doc comment that
CodeRabbit flagged.
- Add TestCreate_MissingTaskDirFields_ReturnsErrTaskDirRequired covering
the three permutations (missing RepoName, missing TaskDirName, both
missing) so a regression in the ErrTaskDirRequired guard surfaces in
CI rather than silently.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
* feat(backend): per-launch authentication for agentctl
Add nonce-based handshake protocol to secure all agentctl HTTP/WS
endpoints. On launch, the parent (backend) passes a one-time bootstrap
nonce via AGENTCTL_BOOTSTRAP_NONCE env var. agentctl generates its own
auth token, enables Bearer auth immediately, and returns the token to
the parent via POST /auth/handshake (validated by nonce).
Standalone, Docker, and Sprites executors all use the same handshake
flow. Removes the old AGENTCTL_AUTH_TOKEN env var approach. For
Docker/Sprites, the retrieved token is persisted encrypted in
SecretStore (AES-256-GCM) for crash recovery.
Security properties:
- Auth enabled from boot (no unprotected window)
- Token never in env vars (only the burned nonce)
- Nonce is one-shot (replay impossible)
- Constant-time token comparison (timing-safe)
* fix: check fmt.Sscanf return value in integration test
* fix: address review feedback - constant-time nonce comparison, safe ID truncation
* fix(backend): exempt MCP routes from agentctl auth middleware
The agent subprocess connects to agentctl's MCP endpoints (/sse, /message,
/mcp) to use kandev tools (create_task, ask_user_question, etc.). These
endpoints run on the same HTTP server as the authenticated API, but the
agent process doesn't possess the auth token — it was only exchanged
between the backend and agentctl during the handshake.
Exempt /sse, /message, and /mcp from the Bearer auth middleware since the
agent process runs in the same trust boundary as agentctl.
* refactor(backend): remove query param token fallback and document SetAuthToken constraint
Remove the ?token= query parameter fallback from extractBearerToken —
tokens in URLs leak via access logs, referer headers, and browser history.
All WS callers already use Authorization headers via wsAuthHeaders().
Document SetAuthToken's concurrency constraint: it must not be called
concurrently with HTTP requests on the same client.
* fix(backend): enforce mutex in ControlClient/Launcher auth, fix Start() lock scope
- Add sync.Mutex to ControlClient; protect SetAuthToken and AuthToken() reads
- Guard Launcher.AuthToken() with l.mu (written under lock in Start, read unprotected)
- Extract buildAndStartProcess and performHandshake from Start() (was 106 lines, over 80-line limit)
- Release l.mu before blocking health/handshake calls so monitorExit() can close
l.exited — previously the fast-fail path in waitForHealthy was unreachable
- Update stale executor_backend.go comment (standalone no longer uses env var auth)
Co-authored-by: João Salavisa <undefined@users.noreply.github.com>
---------
Co-authored-by: Kandev Agent <agent@kandev.dev>
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: João Salavisa <undefined@users.noreply.github.com>
The default Docker API version was hardcoded to 1.41, which fails on
newer Docker daemons that require at least API version 1.44.
Change the default to empty string so the Docker client uses the
existing WithAPIVersionNegotiation() option to auto-negotiate a
compatible version with the daemon.
* feat: replace agent_type with agent_profile_id
- Replace agent_type field with agent_profile_id across backend and frontend
- Add ProfileResolver interface and StoreProfileResolver implementation
- Update task creation/update to use agent_profile_id
- Change default agent runtime from docker to standalone
- Fix workspace state types to include all required fields
- Fix websocket handlers type mismatches for environments, executors, workspaces
- Update integration tests to use agent_profile_id
* feat: pass model as CLI argument to agents
- Add ModelFlag field to AgentTypeConfig for specifying the CLI flag
- Set ModelFlag to '--model' for all agents (auggie, claude, codex, gemini, opencode, copilot)
- Update launchStandalone to append model flag to agent command
- Update launchDocker/buildContainerConfig to append model flag to container command
- Model is passed when both agent supports ModelFlag and profile has Model configured
* fix: make WebSocket message handling async to prevent blocking
When a long-running handler like orchestrator.prompt was executing,
it blocked the ReadPump loop, preventing other requests (like
workspace.tree.get) from being processed until the handler completed.
Now each message is handled in its own goroutine, allowing concurrent
message handling so file tree requests work while the agent is working.
* feat: show task description as first message in chat
- Add 'Task' label with distinct styling (lighter background, border)
- Pass taskDescription from page to TaskChatPanel component
- Differentiate from regular user messages visually
- Add worktree package with Manager, Store, Config, and error types
- Integrate worktree creation/reuse into lifecycle manager
- Update executor to pass worktree configuration
- Add WorktreeConfig to common config package
- Enable worktrees by default for agent isolation
- Add unit tests for worktree manager
This allows multiple agents to work on the same repository simultaneously
by creating isolated worktrees for each task. Worktrees persist between
agent runs for task resumption and are only cleaned up when tasks are
explicitly deleted.
- Use acp-go-sdk for all ACP communication in agentctl
- Mark agent as completed when Prompt() returns successfully
- Update task state to COMPLETED when agent completes
- Update task state to FAILED when agent fails and retries exhausted
- Enhanced E2E test to verify agent actually creates files
- E2E test now validates full flow: task creation -> agent work -> file creation -> task completion
João Salavisa