Add .vscode/ with Biome formatter, Tailwind, Vitest, Playwright, and
Python debug configs. Extract shared pnpm/Node setup into a composite
GitHub Action and add Dependabot and dependency-review workflows.
Convert all 9 AI tool routes (colorize, restore-photo, remove-background,
enhance-faces, blur-faces, red-eye-removal, erase-object, noise-removal,
upscale) to async 202 processing so none are vulnerable to proxy
connection timeouts.
Also fixes:
- Replace basename() with sanitizeFilename() in all AI tool routes
(prevents double-extension attacks and adds length truncation)
- Add UUID format validation for clientJobId field
- Fix missing filename sanitization in noise-removal (was using raw
user-supplied filename with zero sanitization)
- Remove em dash from error message in use-tool-processor
The upscale route held the HTTP connection open for the full duration of
Python sidecar processing (30-300s). Behind proxies with connection
timeouts (Cloudflare Tunnel: 100s), this caused HTTP 524 errors.
The route now returns 202 Accepted immediately after upload validation
and processes in the background. The result (downloadUrl, sizes, etc.)
is delivered via the existing SSE progress channel. The frontend detects
the 202 and waits for the SSE completion event instead of reading the
XHR response body. A reconnect-safe completion store ensures results
survive brief SSE disconnects.
Closes#106
Three interacting bugs prevented GPU inference from ever engaging:
1. onnx_providers() trusted torch.cuda without verifying onnxruntime
actually has CUDAExecutionProvider -- silent CPU fallback
2. Dispatcher close handler counted normal MAX_REQUESTS exits as crashes,
permanently disabling the dispatcher after routine restarts
3. Dispatcher was lazy-started on first AI request with a race condition
that always missed it -- added initDispatcher() for eager startup
Closes#104
Replaces the misleading 'waiting for AI sidecar startup...' message that
never resolved. The dispatcher now starts during server init, and the
startup log shows the actual GPU detection result.
The dispatcher was lazy-initialized on first AI request, but a race
condition meant the first call always missed it (dispatcherReady still
false) and fell through to cold per-request Python. initDispatcher()
starts the dispatcher eagerly and returns a Promise that resolves with
GPU status once ready (or after a timeout).
The close handler called recordCrash() unconditionally, even for exit
code 0 (normal MAX_REQUESTS restart). After 5 normal cycles within 60s
the dispatcher was permanently disabled. Now only non-zero exits count.
gpu.onnx_providers() trusted gpu_available() which returns True via
torch.cuda without checking whether onnxruntime actually has
CUDAExecutionProvider compiled in. When onnxruntime (CPU-only) is
installed, this caused silent fallback to CPU in every ONNX-based tool.
Now verifies onnxruntime.get_available_providers() directly and emits a
diagnostic warning when torch sees CUDA but onnxruntime does not.
Closes#104
Security:
- Apply sanitizeSvg() to all file upload routes (files.ts, user-files.ts)
preventing SSRF and script injection via SVG uploads to file library
Functional:
- Handle PaddleOCR-VL 1.5 markdown_texts output format in ocr.py
- Add empty-text fallback in OCR tier chain (ocr.ts) so higher tiers
that return empty text fall back to the next tier automatically
- Fix SVG->PNG filename extension mismatch in tool-factory.ts so
download endpoint serves correct Content-Type
- Report original upload size (not decoded size) in API response
Test infrastructure:
- Move Playwright auth state from test-results/ to .playwright/ to
prevent mid-run cleanup deleting auth files
- Fix auth.setup.ts navigation race with waitForURL
- Fix gui-batch.spec.ts regex matching "Presets" instead of "reset"
- Fix pipeline-advanced.spec.ts crop bounds and resize assertions
- Broaden pipeline cleanup to include all E2E-prefixed pipelines
captureException now checks isRequestOptedIn before forwarding errors
to Sentry, closing a gap where server errors leaked to an external
service even when no user had consented. The PII scrubbing regex is
also fixed: he[ic]f? failed to match .heic due to word-boundary
behavior and is replaced with hei[cf]? which correctly covers .heic,
.heif, and .hei.
Adds 88 new analytics tests across unit, integration, and e2e layers
proving PostHog/Sentry are never invoked when analytics is disabled or
users have not consented, plus full 7-day reminder lifecycle coverage.
The shouldShowConsent function in the shared package had the correct
logic for checking analyticsConsentRemindAt, but the AuthGuard never
used it. The inline check only redirected to the consent page when
both analyticsEnabled and analyticsConsentShownAt were null, which
is never true after "remind later" since shownAt gets set.
- Destructure analyticsConsentRemindAt from useAuth session
- Hydrate remindAt into the analytics store instead of hardcoding null
- Replace inline redirect check with shouldShowConsent from shared pkg
- Read analyticsConfig from the store inside AuthGuard for serverEnabled
PostHog SDK was initialized on app mount based only on the server-level
config flag, ignoring user consent. This caused network requests to
us-assets.i.posthog.com (config.js, web-vitals.js, dead-clicks-autocapture.js)
even when the user had not opted in or had explicitly declined telemetry.
- Replace static imports of posthog-js and @sentry/react with dynamic
import() so the SDK bundles are not downloaded until consent is granted
- Gate initAnalytics on analyticsConsent.analyticsEnabled === true,
not just server config.enabled
- Add consent re-check after each await import() to handle revocation
during the async load
- Add shutdownAnalytics() that calls opt_out_capturing() + reset()
for mid-session consent revocation
- setAnalyticsConsent(false) now triggers full SDK shutdown automatically
- Rewrite analytics test suite with 44 tests covering init gating,
shutdown lifecycle, consent toggle, race conditions, and Sentry callbacks
Closes#98
Add ~500 new E2E tests and ~300 new integration tests covering:
- 24 new GUI E2E specs: navigation, responsive layout, keyboard shortcuts,
tool UI for all 35 non-AI tools, batch/pipeline workflows, settings/RBAC,
visual regression, accessibility, and performance budgets
- 3 new E2E-Docker specs: batch workflows, advanced pipelines, cross-format
- 1 new adversarial integration test: memory pressure, corrupted files,
unicode filenames, extreme dimensions, pipeline/batch edge cases
- 29 expanded integration test files: HEIC/HEIF input, large files, parameter
boundaries, batch processing, format edge cases across all tools
- Cross-format matrix expanded: 641 tests covering every tool x 18 formats
- AI bridge unit tests expanded: lifecycle, tool modules, error propagation
- Unit test gaps filled: analytics, tool-registry, web stores
Also fixes:
- vitest.config.ts: exclude e2e-docs and e2e-landing from Vitest runner
- AI E2E specs: add sidecar health check to skip gracefully when Python
AI backend is not running instead of timing out
- Add 55 unit tests for feature-status.ts (installed.json CRUD, cache
behavior, install lock, model verification, crash recovery, composite
state) using real temp directories
- Add 36 integration tests for full install/uninstall lifecycle against
Docker containers (face-detection bundle, SSE progress, tool gates,
shared model protection, concurrent install prevention, auth guards,
container restart recovery)
- Fix noise-removal CPU timeout by adding megapixel-based timeout
calculation (120s/MP, min 5 minutes)
- Fix Playwright auth storage state race condition (mkdirSync before
saving analytics-user.json)
- Fix 2 skipped tests in fixes-verification.spec.ts by replacing
external ~/Downloads/sample dependency with existing test fixtures
- Enable skipped analytics-consent settings toggle test
- Restructure features.spec.ts to manage bundle state (uninstall/
reinstall OCR) so 501 guard tests run instead of skipping
- Update noise-removal test mock to include sharp metadata() method
Add 12 tests for background-removal downscaling (resize gate, portrait
orientation, mask upscale) and OOM model fallback (retry with u2net,
progress callback, no-retry guards, cascading failure).
Add OOM propagation tests to face-detection, noise-removal,
red-eye-removal, and OCR -- the four AI features that were missing them.
Skip alpha matting on CPU (pymatting's sparse matrices are the main
memory hog), auto-downscale images above 2048px before sending to
rembg, and retry with the lighter u2net model when OOM is detected.
Main page tool cards and home page tool lists only subscribed to
server-side bundle state, which only reflects the actively downloading
feature. Queued features appeared as plain download icons instead of
showing their queued/installing status. Now subscribes to client-side
installing and queued state from the features store, matching the
settings page behavior.
Register custom BiRefNet-matting ONNX session in install_feature.py so
rembg.new_session("birefnet-matting") no longer raises ValueError during
on-demand installs. The session was already registered in remove_bg.py
(runtime) and download_models.py (build-time) but was missed in the
install path, causing background-removal bundle installs to always fail.
Send JSON body on install/uninstall POST requests to avoid Fastify 5's
strict content-type parser rejecting body-less POSTs with 415.
Fix error message extraction to preserve structured {"error": ...} JSON
from the Python script and filter out pthread_setaffinity_np noise.
HEIF and CLI-decoded formats (DNG, PSD, EXR, HDR, TGA) need external
decoders that are slow on CI runners. Bump timeout from 90s to 180s
to prevent flaky failures. Also extend timeout to CLI-decoded formats
which have the same decode overhead.
Generating a 10000x10000 QR (100 MP) times out even at 120s on
GitHub Actions runners. Use size=2000 instead — boundary validation
(size > 10000 rejected) is already covered by a separate test.
rembg 2.0.62 has both CVE-2026-40086 and GHSA-55v6-g8pm-pw4c
(same vulnerability, different ID sources). Both need ignoring
since upgrading rembg to 2.0.75 breaks the dependency tree.
Pillow 12.x conflicts with pinned numpy 1.26.4, rembg, realesrgan,
and mediapipe. Revert to working 11.1.0 pins and ignore the CVEs
in pip-audit instead — they require a coordinated major version
upgrade across the entire ML stack (Pillow, numpy, torch, basicsr).
Ignored CVEs:
- CVE-2024-27763 (basicsr, no fix available)
- CVE-2026-40086 (rembg, fix needs Pillow 12)
- CVE-2026-25990 (Pillow, fix is 12.1.1)
- CVE-2026-40192 (Pillow, fix is 12.2.0)
- Increase QR generate max-size test timeout to 120s (10000x10000
PNG generation exceeds 30s default on CI runners)
- Update Pillow 11.1.0 → >=12.2.0 (CVE-2026-25990, CVE-2026-40192)
- Update rembg 2.0.62 → >=2.0.75 (CVE-2026-40086)
- Update opencv-python-headless to flexible range >=4.10,<4.12
- Ignore CVE-2024-27763 in pip-audit (basicsr transitive dep from
realesrgan, no fix available upstream)
- Align requirements-gpu.txt and Dockerfile with same versions
- Update OCR engine expected name from "paddleocr" to "paddleocr-v5"
to match actual PaddleOCR PP-OCRv5 engine (eliminates spurious
fallback warning in logs)
- Show "waiting for AI sidecar startup" instead of misleading
"No GPU detected" when Python dispatcher hasn't reported yet
- Fix playwright.docker.config.ts testDir to ./tests/e2e-docker
and align auth storage state path with auth.setup.ts
Closes#17, #18, #19, #31, #32, #33, #34
Format preservation (#17, #18, #19):
- Add resolveOutputFormat to rotate, resize, text-overlay, watermark-text,
border, replace-color, blur-faces, upscale, erase-object, restore-photo
- Alpha-aware fallback: border with corner radius/shadow and replace-color
with makeTransparent fall back to PNG for non-alpha formats (JPEG)
- Python sidecar tools (blur-faces, upscale, erase-object) now convert
PNG output back to input format, matching restore-photo/colorize pattern
- Upscale and erase-object default to "auto" format detection instead of PNG
Dispatcher stability (#31, #32):
- Add gc.collect() and torch.cuda.empty_cache() after each dispatcher request
- Add configurable max_requests (default 50) for periodic dispatcher restart
- Add exponential backoff to dispatcher crash recovery in bridge.ts
- Circuit breaker: 5 crashes within 60s permanently disables dispatcher
- Reset crash counter on successful dispatcher startup
Health & security (#33, #34):
- Export getDispatcherStatus() from @snapotter/ai with running/ready/failed/
gpu/pid/consecutiveCrashes fields
- Admin health endpoint now includes full dispatcher status
- Add pip-audit job to CI workflow for Python dependency scanning
Add djxl (libjxl-tools) as primary JXL decoder with ImageMagick
fallback — fixes JXL format failures on Ubuntu where stock ImageMagick
lacks a JXL delegate. Also make Playwright Docker config respect
BASE_URL env var for testing against remote containers.
Three disconnected systems caused the theme to never apply from server
settings: the DEFAULT_THEME env var was parsed but never seeded to the
database, the settings store ignored defaultTheme from the API, and the
settings dialog wrote to the DB without updating the active theme store.
- Seed DEFAULT_THEME and DEFAULT_LOCALE env vars into the settings table
on first startup (ensureDefaultSettings in index.ts)
- Add applyServerDefault() to theme store that applies the server's
default theme only when the user hasn't made an explicit choice
- Extract defaultTheme from the settings API response and apply it on
fresh sessions (no localStorage preference)
- Apply theme immediately when admin saves settings
- Allow "system" as a valid DEFAULT_THEME env var value
SnapOtter