MOKU 1.0.7

Changelog

All notable changes are documented here. The project follows Semantic Versioning.

[1.0.7] - 2026-07-20

Fixed

• Reject junctions and other reparse points before canonicalization or publication, including those nested below the selected root, and remove every directory created by a failed publish.

[1.0.6] - 2026-07-20

Fixed

• Canonicalize user-selected download roots and final publication paths before containment checks, so equivalent Windows path aliases cannot produce false 502 responses or bypass the save-root boundary.
• Isolate threaded download-integrity test network seams and stabilize the publish-identity test path on hosted Windows runners.

[1.0.5] - 2026-07-18

Added

• Strict multi-tag AND search using ; or ； separators; spaces remain part of one tag.
• Optional bounded anime-oriented tag aliases, disabled by default.
• A collection basket for up to 100 artworks and 1,000 selected images, with windowed page selection for large works.
• One-click select/clear controls for all artworks and images on the current result page.
• Image-first adaptive download chunks and optional artwork grouping.

Changed

• Save one search batch into a shared tag, author, or artwork context folder instead of creating one folder per artwork.
• Apply the same context-folder rule to single-artwork downloads.
• Restyle the interface with a restrained black-and-white lunar theme, a highlighted moon edge, and one clean orbital ring while preserving the existing workflow and startup budget.
• Keep result pagination docked to the viewport bottom while the gallery scrolls.
• Defer the collection retention decision until forward navigation would actually evict selected result pages.

Fixed

• Keep exact and alias-expanded search sessions in separate cache namespaces.
• Validate every multi-tag result against all requested tag groups after Pixiv response normalization.

[1.0.4] - 2026-07-17

Fixed

• Use Pixiv's current JSON /ajax/search/users?nick=... response for exact author: resolution, including both list and keyed user payloads.
• Do not expose the loopback request capability to headerless health probes; same-origin desktop/browser readiness checks now identify themselves explicitly.
• Return download paths relative to the selected save directory instead of leaking local absolute paths through the HTTP API.
• Reject malformed or negative remote Content-Length values before reading a Pixiv response.

[1.0.3] - 2026-07-17

Fixed

• Resolve exact author: queries through Pixiv's current /search/users page instead of the removed AJAX user-search route.
• Parse the bounded __NEXT_DATA__ user result set and keep exact creator-name and user-ID filtering.

[1.0.2] - 2026-07-17

Added

• Exact Pixiv creator search with pid: / pid： and author: / author： queries.

Fixed

• Parse Pixiv's nested userPreviews[].user response before exact author-name matching.
• Reject works whose userId does not match the resolved creator.
• Replace overlapping absolute-positioned deck cards with a non-overlapping flex row.
• Reduce pointer preview travel and keep every other card stationary while one card is locked.

[1.0.1] - 2026-07-16

Fixed

• Restored Windows PowerShell module discovery in -NoProfile CI subprocesses while preserving fail-closed Python Authenticode verification.
• Preserved the vendored proxy-tools license byte hash across Windows and CI checkouts with explicit LF normalization.
• Updated pinned GitHub Actions to Node.js 24-compatible major versions.

[1.0.0] - 2026-07-16

Added

• Windows pywebview/WebView2 desktop host with a second official Pixiv login window.
• Public, R-18, and combined search scopes with multi-tag OR aggregation.
• Bounded historical search, paging prefetch, sliding cache eviction, and selective batch download.
• Native folder selection, offline usage guide, and anonymous parallel network diagnosis.
• Portable PyInstaller build, release ZIP generation, SHA-256 manifests, and Windows CI.
• A fail-closed build manifest binding source/build inputs to every portable-package file.

Security

• Loopback host, Sec-Fetch-Site, and same-origin checks for every API GET.
• Per-process request tokens for every non-health API request; image URLs use separate high-entropy capabilities.
• Strict Content Security Policy and same-origin resource headers on local HTTP responses.
• Bounded JSON-object parsing for mutating requests.
• Explicit Pixiv/API/image host allowlists and loopback-only proxy selection.
• Query parameters, cookies, request bodies, and image tokens excluded from HTTP logs.
• Content-derived backend generation IDs prevent a new client from reusing stale code.
• Test-only synthetic gallery routes disabled by default.
• Release generation rejects stale source, changed licenses, modified support files, linked or undeclared files/directories, non-Windows-x64 product artifacts, missing pywebview loader runtimes, unlocked top-level package metadata, and archives that fail round-trip verification.
• Runtime and build dependencies are locked to verified artifact SHA-256 values; the legacy proxy-tools source is reproduced as an audited deterministic local wheel.
• Build and release validation use a signed CPython 3.12 executable, a shared exclusive mutex, source rechecks, and schema 3 full file/directory manifests.

Changed

• Replaced the legacy external Edge --app host with pywebview/WebView2.
• Replaced sequential port probing with Windows-assigned ephemeral loopback ports.
• Extracted synthetic test fixtures from the production HTTP module.
• Added bounded LRU artwork caching and safe refresh of expired image authorization.
• Synchronized artwork/image-capability state and revalidate in-flight R-18 image/download authorization after network reads.
• Complete staging cleanup before returning download success or failure responses, so the HTTP result matches the final filesystem state.
• Replaced unbounded logs with 5 MiB rotation and removed temporary WebView2 paths from cleanup warnings.
• Removed Android, non-Windows UI backends, unnecessary x86/ARM64 product components, and debug symbols from the Windows x64 frozen closure while retaining the small pywebview loader runtimes required during import.