Review: LexiconForge recent 41 commits#28
Merged
anantham merged 54 commits intoreview/recent-developmentfrom Apr 4, 2026
Merged
Review: LexiconForge recent 41 commits#28anantham merged 54 commits intoreview/recent-developmentfrom
anantham merged 54 commits intoreview/recent-developmentfrom
Conversation
crypto.randomUUID() requires a secure context (HTTPS or localhost). Add a typeof check with Date.now + Math.random fallback so the app loads when accessed via plain HTTP on a network IP.
Add a "Paste Text" tab to InputBar alongside the existing "From URL" tab. Users can paste raw text with a title and source language for reading and translation, bypassing the URL fetch pipeline entirely. The new importCustomText store action constructs an ImportedChapter with a synthetic URL (custom.lexiconforge.local), runs it through the existing transformImportedChapters pipeline, persists to IndexedDB, and merges into the Zustand store — same path as URL-fetched chapters.
The heading, subtitle, and divider were all URL-specific, which doesn't make sense when the Paste Text tab is active. - "OR START FROM A URL" → "OR START FROM SCRATCH" - "Translate Any Web Novel Chapter" → "Translate Any Text" - "Paste a URL from supported sites..." → "Fetch from a supported site or paste your own text"
The default viewMode is 'english', which shows translationResult.translation. Since freshly imported custom text has no translation yet, the content area appeared empty. Now we switch to 'original' view after import so the user immediately sees their pasted text.
MOTIVATION: - Doc audit revealed 45 findings (1 P0, 12 P1, 29 P2, 3 P3) including contradictory file-size policies, stale architecture docs describing completed decompositions as pending, dead links in onboarding, broken test manifest, and orphaned plan files. APPROACH: - Reconcile the three-way file-size policy conflict (AGENTS.md hard limit vs CONVENTIONS.md friction-based vs check-loc.js thresholds) — friction- based policy in CONVENTIONS.md is now authoritative. - Update ARCHITECTURE.md §4.5/§5/§7 to reflect completed decompositions (compiler, navigation, adapters, demoPacket) and fix ghost component references (ParagraphRenderer, IllustrationPlaceholder, etc.). - Migrate active REFACTOR_CANDIDATES watchlist items into ARCHITECTURE.md §7 before archiving; update all referencing docs. - Rebuild TEST_MANIFEST.md from scratch: 122 test files cataloged (was 47 with 74 missing and 13 ghost entries). CHANGES: - AGENTS.md: Fix STOP_CONDITIONS, REQUIRED_READING, bump to v2.1.0 - docs/CONVENTIONS.md: Fix §11 stale entries, add PascalCase divergences - docs/ONBOARDING.md: Fix dead links, align LOC guidance - docs/START_HERE.md: Fix LOC claim, expand ADR index - docs/architecture/ARCHITECTURE.md: Rewrite §4.5, §5, §7, §8 - docs/adr/FEAT-001: Add missing Status header - docs/adr/SUTTA-006: Fix title (was "SUTTA-004"), status → Implemented - docs/features/NOVEL_LIBRARY.md: Document hybrid registry/catalog state - docs/features/IMG2IMG_TESTING.md: Fix steering image path - docs/guides/META_ADAPTER.md: Fix adapter file paths - docs/infrastructure/TEST_MANIFEST.md: Full rebuild from disk scan - docs/roadmaps/README.md: Mark completed/archived items - docs/roadmaps/TECH-DEBT-STATUS.md: Update REFACTOR_CANDIDATES ref - scripts/check-loc.js: Add clarifying comment re: policy authority - Archive: COMPONENT-DECOMPOSITION-PLAN, REFACTOR_CANDIDATES, 3 plans IMPACT: - Agents now receive consistent file-size guidance - New contributors hit no dead links in onboarding flow - Architecture docs accurately reflect current codebase state - Test manifest is a complete, trustworthy inventory TESTING: - All changes are documentation-only (no logic changes) - Verified file moves via git status - TEST_MANIFEST count (122) verified against disk scan Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- SUTTA-003: Status → Implemented (MVP), add implementation notes listing all key files (compiler, UI components, caching, benchmarks) - SUTTA-005: Status → Implemented, add impl notes (leaderboard.json, generate-leaderboard.ts, benchmark UI all confirmed) - FEAT-003: Add implementation notes confirming architecture as-built - ImageGeneration.md: Update steering image list from 4 → 21 (loaded from public/steering-images.json manifest) - STEERING_IMAGE_TEST.md: Archive to diagnostics (one-time test result) - MEMORY_OPTIMIZATION_ROADMAP.md: Verify against code — Phase 1+2 mostly implemented (telemetryService, imageCacheService used by 17 files, imageMigrationService with tests, storage diagnostics UI). LRU eviction (Task 10) and isBlobUrl type flag (Task 3) not adopted. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace derived `hasSession` screen routing with explicit `appScreen` state machine (`library | reader-loading | reader`). Split initializeStore into named phases (settings, boot repairs, deep-link intent, hydration, services). Cold boot always lands on library — no auto-resume. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add libraryScope.ts for consistent novelId::versionId key generation. Schema v15 adds novelId indexes on chapters and url_mappings (also fixes v13 omission). Scoped stableIds prevent cross-version chapter/translation aliasing. ChapterRecord.novelId is now `string | null`. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Single readerHydrationService replaces 4 duplicated hydration paths. ImportService threads registryNovelId and registryVersionId through the full import pipeline. Scoped storage URLs prevent cross-version cache collisions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
BookshelfStateService manages per-novel (per-version) reading position with stale-bookmark fallback chain (lastChapterId -> lastChapterNumber -> chapter 1). Debounced 2s auto-save on chapter change in chaptersSlice. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Continue Reading section shows shelved novels sorted by last read time. Home/Library button in reader header shelves active novel and returns to library. InputBar auto-shelves active novel before URL-paste. NovelCard gains progress badge and resume label. Version picker backed by real storage isolation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Navigation layer respects activeNovelId and libraryVersionId. Scoped cache lookups prevent cross-version bleed — a cache miss for Version A won't silently serve Version B's chapters. Deep links compose sequentially: ?novel imports, then ?chapter navigates. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CLI scripts for building library session files from raw + fan translation sources. Chapter alignment discovery matches chapters across sources by number and content similarity. Independent from the shelf UI. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Update test fixtures to include libraryVersionId on EnhancedChapter and related types. Assertion updates for schema v15. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Record book-switching shelf design spec, implementation plan, and raw-source-discovery spec. WORKLOG entries for version-aware storage work and git cleanup. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Update TECH-DEBT-STATUS with comprehensive audit. Add status headers to FEAT-001 and SUTTA-007 ADRs. Minor fixes to CONVENTIONS, ONBOARDING, START_HERE, ARCHITECTURE, TEST_MANIFEST, and ADDITIONAL-ARCHITECTURAL-ISSUES. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds the ability to switch between library novels without clearing cache. 9 commits: app shell routing, version-aware storage, centralized hydration, bookshelf persistence, library UI with Continue Reading section, navigation version-scoping, importer tooling, test fixtures, and documentation. 974 tests passing. Schema v15. # Conflicts: # docs/WORKLOG.md
Adds Mahāsatipaṭṭhānasutta (MN 10) as a built-in library card that routes to /sutta/demo on click. Appears alongside remote registry novels. Uses the existing interactive Pāli study interface. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add splitPdfParagraphs() heuristic that distinguishes real paragraph breaks from soft wraps in pdftotext output by checking punctuation, capitalization, and line length. - Add rejoinBrokenPdfHeadings() to handle chapter titles split across PDF page/line boundaries. - PDF adapter now uses parseEnglishPdfText() with pdfMode. - Session builder uses fan translation title when available, falling back to raw Chinese title. Before: Chapter 1 had 10 mega-paragraphs (wall of text) After: Chapter 1 has 95 proper paragraphs matching the PDF layout Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- findEnglishBodyStartIndex now checks body length between headings (TOC entries have <500 chars, real chapters have 1000+) - parseEnglishPdfText deduplicates by chapter number, keeping the version with more paragraphs (body > TOC ghost) - Result: 2382 clean chapters from FMC PDF (was 3111 with 730 ghosts) - Only 2 edge-case ghosts remain (chapters 1887, 1915) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The selection popover (thumbs up/down, edit, compare) was gated behind viewMode === 'english', hiding it on the fan translation tab. Changed to only hide on 'original' (raw source text) so users can annotate both fan and AI translations. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The toolbar only listened for mouseup events, which don't fire when users select text via native touch handles. Added selectionchange listener with 200ms debounce to detect touch selections. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- After clicking like/dislike/question, a comment input appears (Save/Skip) - Feedback items render as colored dots in the left gutter near matched text - Click a dot to expand and see the quoted text + comment - Settings toggle: "Show inline feedback markers in left gutter" (default on) - Also: scroll to top on chapter navigation (next/prev) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
npm audit fix resolved all 8 vulnerabilities (rollup, happy-dom, tar, minimatch, picomatch, flatted, brace-expansion, yaml). Project .npmrc adds save-exact, engine-strict, and audit defaults. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… off - Added Part C (Glossary) to default system prompt — starts empty, AI proposes term additions via the existing amendment review flow - Added section 4 to Part A instructing AI when to propose glossary entries - Changed includeFanTranslationInPrompt default from true to false across all providers and settings — fan translation can pollute quality; glossary provides consistency without the noise Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Default debug level changed from 'full' to 'off' - Default pipeline selection returns false (no pipelines) instead of true (all pipelines) when unconfigured - Gated AutoTranslate, App:init, and DefaultKeyBanner logs through debugLog/debugWarn so they respect the settings panel toggles Console is now quiet by default. Enable logging via Settings > Debug. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Cheaper default for new users on the trial key. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Summary-level logs (high-level events like chapter counts, hydration complete, RAM usage) show by default. Full-level verbose spam is suppressed until user enables specific pipelines in Settings > Debug. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…etime Counter now stores the date alongside usage count. If the date doesn't match today, counter resets to 0. Users get 10 free translations per day. Updated banner and error messages accordingly. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two causes: 1. Auto-focusing the input collapses the DOM text selection, which triggers selectionchange → checkSelection → setSelection(null) → popover unmounts. Fixed by skipping selection checks when activeElement is an INPUT/TEXTAREA. 2. Parent div's onMouseDown preventDefault blocks the input from receiving focus on click. Fixed with stopPropagation on the input. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two issues fixed: 1. Grey diff marker explanations (inline "No Change: No explanation provided." text) were captured by window.getSelection() when users selected across paragraphs. Added select-none + aria-hidden so they are excluded from text selections and comparison requests. 2. Comparison card was placed after the selection anchor (first paragraph) instead of the focus (last paragraph). When selecting across two paragraphs, the card appeared between them. Now uses range.endContainer to place the card after the last selected text. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Schema migration v16 adds the api_metrics object store to the centralized DB. Previously apiMetricsService opened its own connection at hardcoded version 14, causing NotFoundError on image generation timing stats. Now uses getConnection() from the shared pool. Converted ~77 raw console.log/console.warn calls to debugLog/debugWarn across 8 files: settingsSlice, openrouterService, importService, navigation/*, apiKeyValidation. All gated by the debug pipeline system — silent by default, visible when user enables the pipeline in Settings > Debug. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
fetchAllNovelMetadata() prepends BUILT_IN_ENTRIES (sutta-mn10) to remote novels. Tests expected counts based only on remote data. Updated 6 assertions to add +1 for the built-in entry. 14/14 tests pass. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Migration test expected 10 stores at version 15. Updated to 11 stores at version 16 after api_metrics addition. 973/974 tests pass (1 pre-existing navigation test failure). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The utils/debug mock only exported debugLog, not debugWarn. When our log gating changes added debugWarn to converters.ts, the missing mock caused a silent TypeError caught by hydration.ts's try-catch, making adaptTranslationRecordToResult appear to return null. All 974 tests now pass. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- AGENTS.md: added DEBT_CAPTURE_PROTOCOL (two-stage debt system) - docs/WORKLOG.md: session notes - .gitignore: added .superpowers/ Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- OpenAI adapter now records `duration` (seconds) in api_metrics for translation calls, enabling empirical time estimates - New getAverageTranslationTime() method with ensemble fallback: exact model → same provider → all translations → 30s default - TranslationCountdownLoader replaces the plain spinner — shows progress bar, estimated time remaining, elapsed time, and sample count from historical data Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a model hits its token output limit (length_cap) and retries with doubled maxOutputTokens also fail, the Translator now automatically splits the chapter into 2-3 chunks at paragraph boundaries and translates each sequentially. Each chunk receives the previous chunk's translation as context for continuity. Results are stitched with merged footnotes (re-numbered), illustrations, usage metrics, and amendments. Also gated all Translator.ts console.log/warn/error calls through debugLog/debugWarn for the translation pipeline. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Flipped enableAmendments default from false to true across all providers, settings UI, store, and promptUtils. The glossary system depends on amendments to grow organically. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Users can now select an image model without triggering automatic illustration generation after every translation. Toggle appears below the image model selector when a model is selected (hidden when model is "none"). Default: on (preserves existing behavior). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Source and target language are book properties, not engine settings. They now display as read-only text in the Translation Engine section with "Edit in Settings > Metadata" hint. Editable only in the Metadata panel where they belong. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two bugs causing all diff markers to be grey no-change: 1. The translation:complete event read fanTranslation from a stale state snapshot captured at the start of handleTranslate. Now reads from live store state via get() to ensure current fanTranslation. 2. Stale cache served results computed without fan translation even after fan translation became available. Added cache invalidation: if cached result has fanHash=null but current request has a real fanHash, the cache is bypassed and diff re-runs with fan text. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Language inputs became read-only text (editable in Metadata only). Tests now check for text content instead of input values/labels. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
To use Codex here, create a Codex account and connect to github. |
Context: the custom text import flow swallowed persistence failures in the store layer, but the paste form still cleared its fields unconditionally.\n\nChanges: wait for importCustomText to return a chapter id before resetting the pasted title, language, and body fields.\n\nImpact: failed custom-text imports now preserve the user\'s pasted content for retry or manual recovery.\n\nTests: /Users/aditya/Documents/Ongoing Local/LexiconForge.worktrees/pr25-fixes/node_modules/.bin/vite build\n\nDocs: none
# Conflicts: # components/InputBar.tsx # store/slices/chaptersSlice.ts
…into codex/pr25-fixes # Conflicts: # components/InputBar.tsx # docs/WORKLOG.md # tests/components/InputBar.test.tsx
…into codex/pr25-fixes
feat: add custom text input via tabbed InputBar
Owner
Author
|
Review-only PR — closing, all code is on main |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Retroactive review of 41 recent commits including:
@codex please review
🤖 Generated with Claude Code