FastAPI auto-redirects /workspaces to /workspaces/ with 307.
Behind Traefik (HTTP internal), the 307 becomes http://,
triggering Mixed Content in the browser. Adding trailing
slashes avoids the redirect entirely.
Fixes NameError: ToolInstance not defined at runtime because
type annotations are evaluated at class definition time.
Deferring annotation evaluation with __future__ annotations
keeps TYPE_CHECKING imports from causing runtime crashes.
Also includes ruff formatting cleanup on workspace-related files.
The API container used a named Docker volume (repo_data:/data/repos) for
storing repositories. When creating tool instances with direct mount mode,
the API told Docker to bind-mount /data/repos/<repo>:/workspace into the
tool container. But the Docker daemon resolves bind-mount paths on the HOST
filesystem, not inside the API container. Since the host had no /data/repos
(the repos only existed inside the named volume), tool containers mounted
empty directories.
Changed both compose files to use a host bind mount (/data/repos:/data/repos)
instead of a named volume. This ensures:
- The API container and tool containers both see the same /data/repos path
- Bind mounts from /data/repos into tool containers work correctly
For existing installations: repos previously stored in the repo_data named
volume should be copied to /data/repos on the host before restarting the
stack.
Quality gates: compose file syntax valid
- apps/api/src/services/tunnel.py: add 2-second sleep after discovering the
tunnel URL to allow Cloudflare DNS edge propagation before returning
- apps/web/src/hooks/use-instance-actions.ts: show alert() with the backend
error message when recreate tunnel fails, instead of silently swallowing
errors
Quality gates: ruff clean, tsc clean
Replace the subprocess-based tunnel implementation with Docker containers
running on the host network. This eliminates all container name resolution
bugs that caused tunnel 502 errors.
New design:
- Each tunnel is a docker run --network host cloudflare/cloudflared container
- cloudflared connects to localhost:{published_port} (Docker port forwarding)
- No dependency on container names, backend network DNS, or binding diagnostics
- Tunnels named predictably: tunnel-{instance_name}
- Start/stop/recreate use container names instead of PIDs
Files changed:
- NEW: apps/api/src/services/tunnel.py — clean tunnel module (start/stop/recreate/health)
- apps/api/src/services/docker.py — removed 250 lines of old tunnel code
- apps/api/src/api/tool_instances.py — use new tunnel module, store container_name
- apps/api/src/services/health_monitor.py — updated import
- apps/web/src/components/session-card.tsx — Recreate Tunnel button always visible
Quality gates: ruff clean, 13 tests passed (health_monitor + notifications)
Show the Recreate Tunnel button on all active web-enabled session cards
(instead of only when tunnel_status is unreachable) so users can manually
trigger tunnel recreation at any time. Also adds it to the mobile action sheet.
Quality gates: eslint clean, tsc clean
- Dual-mode touch scroll:
- Normal mode: scroll .xterm-viewport directly when scrollHeight > clientHeight
- Alternate screen (tmux/vim): send SGR 1006 mouse-wheel protocol data
using cursor position so tmux knows which pane to scroll
- Add touch-action: none to .terminal-container to prevent browser gestures
- Lock both html and body overflow when terminal page is open on mobile
- Remove synthetic WheelEvent approach (xterm.js SmoothScrollableElement
doesn't reliably handle synthetic events)
- Attach capture-phase touch listeners to .terminal-container (parent of xterm)
- On vertical swipe: e.preventDefault() blocks page scroll, then directly
adjust .xterm-viewport.scrollTop by the swipe delta
- This bypasses term.scrollLines() API and directly manipulates the DOM
element that xterm.js watches via its internal scroll handler
- Remove all CSS touch-action overrides — container handles it in JS
- Add full mobile viewport CSS: overflow-y scroll, -webkit-overflow-scrolling
touch, overscroll-behavior-y contain, translate3d hardware accel,
scroll-behavior smooth, touch-action pan-y
- After term.open(), find .xterm-viewport and add passive touch listeners
that call stopPropagation() (not preventDefault) — this lets the browser
handle native touch scrolling while preventing xterm.js internal handlers
from interfering
- Based on xterm.js known issue #5489 and SCROLLING_FIX.md approach
- xterm.js has zero touch event handlers (verified: only 1 'touch' ref in
entire library), so it wasn't intercepting anything
- Our touch-action: none + preventDefault() combo was blocking the browser
from scrolling the .xterm-viewport natively
- Removed all custom touch event handlers from terminal.tsx
- Removed touch-action: none from .terminal-container
- Added touch-action: pan-y to .xterm-viewport so browser allows vertical pan
- Body scroll lock (terminal-page-open) prevents page from scrolling
- Attach touch listeners to document with capture:true instead of container
- Check if touch target is inside terminal container before handling
- This runs before xterm.js internal handlers, giving us full control
- Add touch-action: none to terminal container to prevent browser gestures
- Lower threshold to 3px, 20px per line for responsive scrolling
- Add body.terminal-page-open { overflow: hidden } to prevent page scroll
- TerminalPage adds/removes 'terminal-page-open' class on body when mounted
- Re-add capture-phase touch listeners in terminal.tsx with low 3px threshold
- Call e.preventDefault() immediately when vertical gesture is detected,
before browser compositor commits to page scroll
- Remove CSS touch-action overrides on xterm viewport (now handled in JS)
- Scroll forwarded via term.scrollLines() with 24px per line sensitivity
- Remove all custom touch event interception code from terminal.tsx
- After term.open(), find the internal .xterm-viewport element and set
touchAction=pan-y and overscrollBehavior=contain via inline styles
- Add CSS targeting .xterm-viewport on mobile with touch-action: pan-y,
-webkit-overflow-scrolling: touch, and overflow-y: auto
- Let the browser handle vertical touch panning natively instead of
trying to intercept and manually forward events
- Attach touch listeners to container wrapper in CAPTURE phase so they run
before xterm.js internals stop propagation
- Add e.stopPropagation() in touchmove after handling scroll to prevent
xterm.js from conflicting with our scroll
- Add wheel event fallback for mobile browsers that synthesize wheel from touch
- Remove touch-action: none CSS which was blocking native xterm viewport scroll
- Attach touch listeners to term.element (xterm root) instead of wrapper
- Fix scroll direction: swipe up now scrolls up (shows older buffer)
- Remove RAF indirection; scroll applied synchronously in touchmove
- Accumulate delta between events for smoother scrolling
- Lower threshold to 6px and px-per-line to 16 for better responsiveness
- Add touch-action: none to terminal container on mobile
- Intercepts touch events on the terminal container when isMobile=true
- Detects vertical swipe gestures (dominant over horizontal movement)
- Translates swipe distance to xterm.js scrollLines() calls
- Uses requestAnimationFrame for smooth scroll updates
- Threshold of 10px before scroll kicks in; 30px per line
- Touch listeners cleaned up on component unmount
- Uses navigator.wakeLock.request('screen') to keep device awake
- Re-acquires wake lock when tab becomes visible again
- Releases wake lock on component unmount
- Silently ignored on unsupported browsers or if denied
- Add SpecialKeysStrip and SpecialKeysPanel to mobile terminal page
- Store sendData and focusInput refs via onTerminalReady callback
- Pass activeModifier/onModifierChange to TerminalComponent on mobile
- Add virtual keyboard padding to prevent keyboard from covering terminal
- Special keys bar sits at bottom of viewport, panel opens as overlay
- Replace inline header+tabs layout with position:absolute overlay
- Overlay contains: back button, session name, status dot, A-/A+ font size, exit
- Session tabs live inside the overlay below the toolbar
- Auto-hides after 3s; clicking terminal content hides it immediately
- Pull handle at top edge appears when overlay is hidden to restore it
- Terminal content always fills full viewport; overlay never resizes container
- Pass showControls=false to TerminalComponent on mobile to avoid double headers
Notification filtering:
- lifecycle_hooks.py: only instance.error and instance.health_changed
with status=running generate notifications. All other lifecycle events
(created, started, stopped, restarted, deleted) are filtered out.
- health_monitor.py: only error and unhealthy states generate notifications.
Running/recovered state no longer creates info notifications.
- _derive_title now maps instance.health_changed to "Container ready".
Clear-all button:
- Added dismiss_all() to NotificationService
- Added DELETE /notifications endpoint for bulk dismiss
- Frontend: clearAllNotifications API, clearAll in notification context,
"Clear all" button in notification drawer alongside "Mark all as read"
- Added CSS for .notification-clear-all with danger hover state
- Updated notification-center tests
Quality gates: pytest (21 passed), vitest (11 passed)
The create-session-form was calling startInstance() without passing the
selected config profile and SSH keys. This caused the backend to receive
ssh_key_ids=[] and clear the keys that were stored during createInstance.
The .ssh directory was never mounted because instance.ssh_key_ids was
wiped during the start call.
Also includes minor formatting cleanup on the data migration.
Quality gates: pytest (18 passed)
- Add ssh_key_id column to ConfigProfile model and migration
- Update config profile API to accept/return ssh_key_id
- Include ssh_key_id in ResolvedProfile and resolver logic
- Mount selected SSH key into container home dir at start_instance
- Frontend config profile form with SSH key selector dropdown
- Git mount URL validation defaults to profile's SSH key
Quality gates: pytest (231 passed, 6 pre-existing), tsc --noEmit clean
- Add POST /config-profiles/validate-git-url endpoint:
- Parses URL using existing parse_git_url utility
- Suggests corrected URL for browser URLs
- Runs git ls-remote --heads to verify reachability
- Lists available branches from remote
- Supports SSH key for private repos
- Returns structured response: valid, suggested_url, branches,
default_branch, error, error_code
- Update frontend GitMountEditor:
- Add Check button next to URL field with loading state
- Show validation result: valid (green), suggestion (yellow),
invalid (red)
- Suggestion includes Use this button to apply corrected URL
- Branch field becomes dropdown when URL is validated,
populated with remote branches
- Mappings section disabled until URL is validated
- Shows hint: Validate the URL first
- Quality gates: pytest (218 passed, 6 pre-existing),
tsc --noEmit (clean)
- Remove global Escape key listener that intercepted Escape before
xterm.js could receive it, breaking vim/tmux/etc.
- Add click-outside-to-exit for fullscreen: clicking on the padding
area around .terminal-page-content or .terminal-fullscreen-header
exits fullscreen. Clicks inside content or header are ignored.
- Add 8px padding/gap to .terminal-page.fullscreen to create a
clickable border area around the terminal.
- Keep Exit button and Alt+Shift+F as explicit exit methods.
Quality gates: tsc --noEmit (clean), pytest (208 passed, 6 pre-existing)
When switching terminal sessions in fullscreen mode, the viewport
shrank cumulatively because .terminal-wrapper uses
grid-template-rows: auto 1fr. With showControls=false, the single
child (.terminal-container) landed in the auto track instead of 1fr,
creating a feedback loop with xterm fit().
- Add .terminal-wrapper.no-controls with grid-template-rows: 1fr
so the container fills the wrapper when the header is hidden.
- Apply no-controls class in TerminalComponent when showControls=false.
- Replace setTimeout(50) with double requestAnimationFrame in
TerminalPage for more reliable fit() timing after tab switches.
Quality gates: tsc --noEmit (clean), pytest (208 passed, 6 pre-existing)
- Add expand_container_path() helper that resolves ~/ and $HOME/ prefixes
- Add get_manifest_home_dir() to compute /home/{user.name} or /root from manifest
- Set ENV HOME=... and ENV USER=... in generated Dockerfile for runtime compatibility
- Pass home_dir through instance creation and startup pipeline
- Expand mount targets in apply_resolved_profile() for regular profile mounts
- Expand mapping targets in _resolve_git_mount_mappings() for git mounts
- Expand working_directory and volume targets in _modify_compose_file()
- Update _prepare_manifest_instance to return home_dir alongside image tag
- Fetch tool_type early in start_instance to determine home_dir before profile application
Quality gates: pytest 188 passed, frontend typecheck clean
Addresses: home-path-expansion
- Add mappings array support to git_mount entries
- Clone repository once per git_mount entry, mount multiple subdirectories
- Normalize legacy source_path+target_path to mappings on read
- Update _merge_git_mounts to dedup by (remote_url, branch) and concatenate mappings
- Add _normalize_git_mount, _clone_git_repo, _resolve_git_mount_mappings helpers
- Update GitMountItem Pydantic model with GitMountMapping and model_validator
- Update frontend GitMountEditor component with mappings UI
- Auto-convert legacy git mount entries to mappings format on load
- Add 15 backend unit tests for normalization, resolution, and glob expansion
- Update existing config profile resolver tests for new merge behavior
Quality gates: pytest 167 passed, frontend typecheck clean
Addresses: config-profile-multi-repo-mounts
Frontend:
- Remove 4 console.log statements from terminal.tsx that flooded the
browser console with WebSocket traffic (open, received X bytes, sending Y,
xterm focused)
Backend:
- Downgrade Dockerfile/entrypoint compilation logs from INFO to DEBUG in
_prepare_manifest_instance
- Remove hex-dump diagnostic logging from docker_build.py (was for
troubleshooting the backslash continuation bug, now fixed)
- Downgrade Dockerfile write log from INFO to DEBUG
The ManifestEditor had a feedback loop:
1. State change → buildManifest changes → onChange notifies parent
2. Parent updates manifestData → new manifest prop
3. Loading effect sets all state from manifest (arrays get new refs even if same content)
4. New array refs → buildManifest changes → onChange fires again → loop
Fix: track the last-sent manifest via a ref and only call onChange when the
serialized built manifest actually differs. This breaks the cycle because after
the loading effect syncs state, the rebuilt manifest is identical in content
so we skip the parent notification.
Backend:
- Allow 'manifest' in tool_types definition_type validators
- Add manifest_id to ToolTypeCreate, ToolTypeUpdate, ToolTypeResponse
- Skip compose/dockerfile template validation when definition_type is manifest
- Require manifest_id when definition_type is manifest
- Clear legacy templates when switching to manifest type
Frontend:
- Load manifest data via getToolDefinition when selecting a manifest-type tool
- Create/update manifest definition via tool-definitions API when saving
- Pass manifest_id to tool-types create/update API
- Fix unused EmptyState import after configs/folders cleanup
The instance status may say 'running' but the actual Docker container
may have been removed (e.g. docker prune, host restart). The old code
created a terminal session which immediately died because docker exec
failed with 'No such container'.
- Add get_container_status check in WebSocket handler before session creation
- Return 4004 with clear message if container is missing
- This prevents spawning zombie terminal sessions
- Initialize loading=true in useTerminalSessions to prevent auto-create
from firing before initial load completes
- Remove hasAutoCreated ref from TerminalPage (no longer needed)
- Add focus() to TerminalRef, call on tab switch
- Add term.focus() after term.open() in TerminalComponent
- Add console logging for WebSocket send/receive to debug no-i/o
- Revert backend _read_loop retry logic to original break-on-error
Three fixes for multi-session terminal bugs:
1. Race-condition double creation: The auto-create effect fired twice because
loadSessions returned 0 while an earlier createSession was still in flight.
Added hasAutoCreated guard ref to ensure only one auto-create happens.
2. Page reload spawns new sessions: After server restart, list_terminal_sessions
filtered out DB-only sessions (no in-memory counterpart), so the frontend
thought no sessions existed and auto-created new ones. Reverted the filter
so DB rows are always returned. The WebSocket handler now restores the
in-memory session from the DB row on demand when connecting.
3. No input after connection: The backend _read_loop would break on any send
error, causing asyncio.wait to cancel the _write_loop. Made _read_loop
retry up to 3 times before giving up, preventing transient send errors
from killing input handling.
Quality gates: pytest (15/15 passed), tsc clean
Three related bugs fixed:
1. Frontend xterm.js crash: TerminalPage rendered ALL sessions with display:none
for inactive ones. xterm.js crashes when initialized in a hidden container
(Viewport can't read dimensions). Fix: only render the active session's
TerminalComponent using conditional rendering.
2. Backend websocket disconnect cascade: When client disconnected (due to #1),
the server tried to send 'connected' status on dead socket, caught the
WebSocketDisconnect in a generic except block, then tried to close() again
causing RuntimeError. Fix: catch WebSocketDisconnect specifically and suppress
close() errors.
3. Stale DB sessions: After server restart, DB still had old terminal session
rows but no in-memory sessions. list_terminal_sessions returned these ghosts,
causing the frontend to render dead tabs. Fix: skip DB-only sessions that
have no live in-memory counterpart.
Quality gates: pytest (15/15 passed), tsc clean, vitest (7/7 passed)
alex
Alex Blank