Add 5 WebGL shader background effects (Aurora, Plasma, Digital Rain,
Starfield, Warp Tunnel) via a new bg-shaders.ts engine that shares
a dedicated canvas. Add 5 style presets (Sakura, Ocean, Copper, Vapor,
Monolith) with distinctive font pairings. Remove CSS particles effect
in favor of shader-based alternatives. Fix dot grid visibility and
tune all shader intensities for subtle ambient appearance.
Add style presets (font + color combinations) and background effect
presets as a new Appearance tab in the settings modal. Style presets
include Default, Midnight, Ember, Arctic, Terminal, and Neon — each
with curated dark/light theme colors and Google Font pairings.
Background effects (Dot Grid, Gradient Mesh, Scanlines, Particles)
use a dedicated overlay div alongside the existing WebGL Noise Field.
All choices persist to localStorage and restore on page load.
Scene name and fallback scene in automation cards are now clickable,
navigating to the corresponding scene preset card. Also renders the
deactivation mode label which was previously set but never displayed.
Auth is now optional: when `auth.api_keys` is empty, all endpoints are
open (no login screen, no Bearer tokens). Health endpoint reports
`auth_required` so the frontend knows which mode to use.
Backup/restore fixes:
- Auto-backup uses atomic writes (was `write_text`, risked corruption)
- Startup backup skipped if recent backup exists (<5 min cooldown),
preventing rapid restarts from rotating out good backups
- Restore rejects all-empty backups to prevent accidental data wipes
- Store saves frozen after restore to prevent stale in-memory data
from overwriting freshly-restored files before restart completes
- Missing stores during restore logged as warnings
- STORE_MAP completeness verified at startup against StorageConfig
Camera engine and video stream support now gracefully degrade when
opencv-python-headless is not installed. The app starts fine without
it — camera engine simply doesn't register and video streams raise
a clear ImportError with install instructions.
Saves ~45MB for users who don't need camera/video capture.
- Notification history: replace text buttons with icon buttons, use modal-footer for proper positioning
- CSS test: reject 0-LED picture sources with clear error message, show WS close reason in UI
- Calibration: distribute LEDs by aspect ratio (16:9 default) instead of evenly across edges
- Value source schedule: replace native time input with custom HH:MM picker matching automation style
- Remove "No ... yet" empty state labels from all CardSection instances
Create contexts/ci-cd.md documenting release pipeline, build scripts,
CI runners, and versioning. Reference it from CLAUDE.md context table.
Add mandatory pre-commit lint check rule to CLAUDE.md.
Add Windows installer, Docker volume mount, and first-time setup
instructions to the Gitea release body. Fix Docker registry URL.
Add CI/Release sync rule to CLAUDE.md.
Add compileall step to build-dist-windows.sh that generates .pyc files
for both app source and site-packages. Saves ~100-200ms on startup by
skipping parse/compile on first import.
- Replace winsdk (~35MB) with winrt packages (~2.5MB) for OS notification
listener. API is identical, 93% size reduction.
- Replace wmi (~3-5MB) with ctypes for monitor names (EnumDisplayDevicesW)
and camera names (SetupAPI). Zero external dependency.
- Migrate cv2.resize/imencode/LUT to Pillow/numpy in 5 files (filters,
preview helpers, kc_target_processor). OpenCV only needed for camera
and video stream now.
- Fix DefWindowProcW ctypes overflow on 64-bit Python (pre-existing bug
in platform_detector display power listener).
- Fix openLightbox import in streams-capture-templates.ts (was using
broken window cast instead of direct import).
- Add mandatory data migration policy to CLAUDE.md after silent data
loss incident from storage file rename without migration.
Strip unnecessary files from site-packages:
- Remove pip, setuptools, pythonwin (not needed at runtime)
- OpenCV: remove unused extra modules and data
- numpy: remove tests, f2py, typing stubs
- Remove all .dist-info directories and .pyi type stubs
- Remove winsdk type stubs
- Replace 7z with msiextract (msitools) to extract tkinter from
python.org's individual MSI packages (tcltk.msi + lib.msi)
- Fix build step numbering to /9
- Docker job continues on login failure (registry may not be enabled)
- Show makensis output for debugging
- Replace nuget approach (doesn't contain tkinter) with extracting
from the official Python amd64.exe installer using 7z
- Remove MUI_ICON/MUI_UNICON (no .ico file available, use NSIS default)
- Add p7zip-full to CI dependencies
- installer.nsi: per-user install to AppData, Start Menu shortcuts,
optional desktop shortcut and autostart, clean uninstall (preserves
data/), Add/Remove Programs registration
- build-dist-windows.sh: runs makensis after ZIP if available
- release.yml: install nsis in CI, upload both ZIP and setup.exe
- Fix Docker registry login (sed -E for https:// stripping)
Download _tkinter.pyd, tkinter package, and Tcl/Tk DLLs from the
official Python nuget package and copy them into the embedded Python
directory. This enables the screen overlay visualization during
calibration in the portable build.
- Move startup banner into main.py so it shows the actual configured
port instead of a hardcoded 8080 in the launcher scripts
- Wrap tkinter import in try/except so embedded Python (which lacks
tkinter) logs a warning instead of crashing the overlay thread
Windows: install-autostart.bat (Startup folder shortcut),
uninstall-autostart.bat. Linux: install-service.sh (systemd unit),
uninstall-service.sh.
Both launchers now use python -m wled_controller.main so port is
read from config/env instead of being hardcoded to 8080.
Windows embedded Python ignores PYTHONPATH when a ._pth file exists.
Add ../app/src to the ._pth so wled_controller is importable.
Fixes ModuleNotFoundError on portable builds.
Replace Windows runner requirement with cross-compilation:
download Windows embedded Python + win_amd64 wheels from PyPI,
package into the same ZIP structure as build-dist.ps1.
All 4 release jobs now run on ubuntu-latest.
Restructure release.yml into 4 jobs:
- create-release: shared Gitea release with download table
- build-windows: existing portable ZIP (unchanged)
- build-linux: new tarball with venv + run.sh launcher
- build-docker: push image to Gitea Container Registry
Add build-dist.sh as Linux equivalent of build-dist.ps1.
Docker tags: version + latest (stable only, no latest for alpha/beta/rc).
Tests that call get_available_displays() or capture_display() require
a real display ($DISPLAY on Linux, always available on Windows/macOS).
Mark them with @requires_display to skip on headless CI instead of
failing with "$DISPLAY not set".
Affects: test_screen_capture.py (4 tests), test_api.py (1 test).
- Lazy-import tkinter in screen_overlay.py (TYPE_CHECKING + runtime
import) so the module loads on headless Linux CI without libtk8.6
- Fix test_wled_client.py: mock all HTTP endpoints with respx (info,
cfg, state) instead of hitting real network
- Fix test_calibration.py: assert numpy array shape instead of tuple
- Fix test_processor_manager.py: update to current API (async
remove_device, dict settings, no update_calibration)
- Fix test_screen_capture.py: get_edge_segments allows more segments
than pixels
341 tests passing, 0 failures.
screen_overlay.py imported tkinter at module level, which cascaded
through processor_manager → every test touching the app on CI where
libtk8.6.so is unavailable. Move to TYPE_CHECKING + runtime lazy
import so the overlay module loads cleanly on headless systems.
Also fix test_processor_manager.py to use ProcessorDependencies().
- Refactor process-picker.ts into generic NamePalette with two concrete
instances: ProcessPalette (running processes) and NotificationAppPalette
(OS notification history apps)
- Notification color strip app colors and filter list now use
NotificationAppPalette (shows display names like "Telegram" instead of
process names like "telegram.exe")
- Fix case-insensitive matching for app_colors and app_filter_list in
notification_stream.py
- Compact browse/remove buttons in notification app color rows with
proper search icon
- Remove old inline process-picker HTML/CSS (replaced by palette overlay)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Server: send initial frame immediately after metadata for api_input
sources so the client gets the current buffer (fallback color if
inactive) instead of never receiving a frame when push_generation
stays at 0.
Frontend: clear all test modal canvases on open to prevent stale
pixels from previous sessions. Reset FPS timestamps after metadata
so the initial bootstrap frame isn't counted in the FPS chart.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace the body position:fixed hack with overflow:hidden on html element,
which works cleanly with scrollbar-gutter:stable to prevent layout shift.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Rename all 54 .js files to .ts, update esbuild entry point
- Add tsconfig.json, TypeScript devDependency, typecheck script
- Create types.ts with 25+ interfaces matching backend Pydantic schemas
(Device, OutputTarget, ColorStripSource, PatternTemplate, ValueSource,
AudioSource, PictureSource, ScenePreset, SyncClock, Automation, etc.)
- Make DataCache generic (DataCache<T>) with typed state instances
- Type all state variables in state.ts with proper entity types
- Type all create*Card functions with proper entity interfaces
- Type all function parameters and return types across all 54 files
- Type core component constructors (CardSection, IconSelect, EntitySelect,
FilterList, TagInput, TreeNav, Modal) with exported option interfaces
- Add comprehensive global.d.ts for window function declarations
- Type fetchWithAuth with FetchAuthOpts interface
- Remove all (window as any) casts in favor of global.d.ts declarations
- Zero tsc errors, esbuild bundle unchanged
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fix indentation bug causing scenes device to not register
- Use nonlocal tracking to prevent infinite reload loops on target/scene changes
- Guard WS start/stop to avoid redundant connections
- Parallel device brightness fetching via asyncio.gather
- Route set_leds service to correct coordinator by source ID
- Remove stale pattern cache, reuse single timeout object
- Fix translations structure for light/select entities
- Unregister service when last config entry unloaded
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Bulk selection mode: Ctrl+Click or toggle button to enter, Escape to exit
- Shift+Click for range select, bottom toolbar with SVG icon action buttons
- All CardSections wired with bulk actions: Delete everywhere,
Start/Stop for targets, Enable/Disable for automations
- Remove expand/collapse all buttons (no collapsible sections remain)
- Fix graph node color picker overlay persisting after outside click
- Add Icons section to frontend.md conventions
- Add trash2, listChecks, circleOff icons to icon system
- Backend: processing loop performance improvements (monotonic timestamps,
deque-based FPS tracking)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Previously modals showed generic "Failed to add/save" messages. Now they
extract and display the actual error detail from the API response (e.g.,
"URL is required", "Name already exists"). Handles Pydantic validation
arrays by joining msg fields.
Fixed in 8 files: device-discovery, devices, calibration,
advanced-calibration, scene-presets, automations, command-palette.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- KC test WS now acquires from LiveStreamManager instead of creating its
own DXGI duplicator, eliminating capture contention with running LED targets
- Tree-nav refactored to compact dropdown on click with outside-click dismiss
(closes on click outside the trigger+panel, not just outside the container)
- KC target card badge (e.g. "Daylight Cycle") no longer wastes empty space
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Use { behavior: 'instant' } in unlockBody scrollTo to override
CSS scroll-behavior: smooth on html element
- Use { preventScroll: true } on focus() restore in Modal.forceClose
- Add overflow-y: scroll to body.modal-open to prevent scrollbar shift
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fix CSS specificity: .dimmed now overrides .graph-edge-active opacity/filter
- Add data-from/data-to to flow dot groups so they can be dimmed per-edge
- Dim flow dots on non-chain edges in highlightChain(), restore on clear
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Settings modal split into 3 tabs: General, Backup, MQTT
- Log viewer moved to full-screen overlay with compact toolbar
- External URL setting: API endpoints + UI for configuring server domain
used in webhook/WS URLs instead of auto-detected local IP
- Sources tab tree restructured: Picture Source (Screen Capture/Static/
Processed sub-groups), Color Strip, Audio, Utility
- TreeNav extended to support nested groups (3-level tree)
- Audio tab split into Sources and Templates sub-tabs
- Fix audio template test: device picker now filters by engine type
(was showing WASAPI indices for sounddevice templates)
- Audio template test device picker disabled during active test
- Rename "Input Source" to "Source" in CSS test preview (en/ru/zh)
- Fix i18n: log filter/level items deferred to avoid stale t() calls
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Graph editor:
- Floating FPS tooltip on hover over running output_target nodes (300ms delay)
- Shows errors, uptime, and FPS sparkline seeded from server metrics history
- Tooltip positioned below node with fade-in/out animation
- Uses pointerover/pointerout with relatedTarget check to prevent flicker
- Fixed-width tooltip (200px) with monospace values to prevent layout shift
- Node titles show full names (removed truncate), no native SVG <title> tooltips
Documentation:
- Added duration/numeric formatting conventions to contexts/frontend.md
- Added node hover tooltip docs to contexts/graph-editor.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New features:
- Override composite blend mode: per-pixel alpha from brightness
(black=transparent, bright=opaque). Ideal for API input over effects.
- API input test preview FPS chart uses shared createFpsSparkline
(same look as target card charts)
Fixes:
- Fix api_input source not surviving server restart: from_dict was
still passing removed led_count field to constructor
- Fix composite layer brightness/processing selectors not aligned:
labels get fixed width, selects fill remaining space
- Fix CSPT input selector showing in non-CSPT CSS test mode
- Fix test modal LED/FPS controls showing for api_input sources
- Server only sends test WS frames when api_input push_generation changes
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When HA sets a color via turn_on, also update the source's fallback_color
to match. This way when the api_input timeout fires (default 5s), the
stream reverts to the same color instead of black. turn_off resets
fallback to [0,0,0].
Added coordinator.update_source() for PUT /color-strip-sources/{id}.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
