270 Merged PRs in GenOffice: How I Hardened an Open-Source AI Office Suite

From single-quoted XML quirks to non-finite geometry guards — the story behind 270 merged pull requests in genspark-ai/genoffice, the world's first full-featured open-source AI Office suite.

GenOffice is a free, open-source alternative to Microsoft Office (macOS / Windows / Linux). It opens and saves native .docx, .xlsx, .pptx, edits PDF, Markdown and HTML, and puts a reviewable AI agent next to every document. Real formats, byte-preserving. Local by design.

I contributed 270 merged PRs (PR #131 → #733), backed by ~268+ commits in my local clone. Almost all are fix: / test: — small, reviewable, CI-green hardening patches across Docs, Sheets, Slides, PDF, DOCX engine, PPTX engine, xlsx-gateway, html2docx, pdf2docx, shell, CLI, electron-utils, ai-provider, ai-search, and a11y/i18n.

Repo: https://github.com/genspark-ai/genoffice

How it started

I came for the file-format rabbit hole and stayed for the edge cases. Office files are hostile inputs in the wild: single-quoted attributes, uppercase True, paired protection tags, BOM-less UTF-16 CSVs, backslash pptx targets, lowercase xlsx refs, broken slide sizes, 1904 date systems. Each one breaks a naive parser. My loop became: reproduce → clamp/validate/normalize → add regression test → keep the diff minimal.

Early work (#131–#160) was Sheets/Slides streaming and save reliability: Ctrl+F beyond the loaded window, filter-hidden rows in find, whole-file error checking on streamed workbooks, row-batch sidecar reads, temp-file TTL sweeps, serialized concurrent saves, in-flight save waits, SSE reader release + malformed-frame skips, IPC failure surfacing.

What I actually did — by theme

1. Parser hardening: read what Word really writes

• DOCX: single-quoted fldCharType, protection attrs, checkbox glyphs, footnote separators, TOC tab stops, xml:space, numeric char refs, NBSP edges, paired tags, vanish/footnote bold variants.
• PPTX: single-quoted slide ids / hyperlink r:id / showMasterSp, preset colors case-insensitive, true/True table flags, tab stops as tabs, slides in presentation order.
• XLSX / CSV: lowercase cell refs, escaped quotes + quote-state across lines, BOM-less UTF-16, over-precision integers as text, delimiter sniffing that ignores single-column prose.

2. Numeric safety: kill NaN / Infinity / overflow

A huge slice is one pattern: non-finite in → sane out. Chart axis min/max, font sizes, ink geometry, connector widths, border widths, rect dims, table colSpan, EQ columns, slide-id scans, CSV row/col caps, bitmap dims, outline JSON caps, SSE buffers, tool-call counts, chat limits, timeline limits, print geometry, AI length parsers (parseTableLength #730). Boring? Yes. That's why it matters — one NaN can DoS a renderer or corrupt an export.

3. Security + IPC + SSRF + save reliability

Traversal-id rejection, canonical path comparison for tabs, renderer file resolution hardening, SSRF maxRedirects normalization, MCP host/file checks, sheet-address hardening, zip-handle closing + central-directory validation, protection spinCount clamping, lazy-media hash/part-path hardening, redaction-region caps, outline-depth caps, link-rect validation, serialized saves (#155), in-flight save waits (#149), SSE cleanup (#150/#151).

4. i18n / a11y / RTL

Bidi direction tests, Southeast Asian + CJK word counts, kana/radical/astral-CJK font mapping, hardcoded cm strings moved into i18n (#733), existing i18n keys for aria labels, dialog role/focus/escape, diagram preview keyboard operable, toasts + outline empty-state announcements, localized labels, RTL-aware AI panels.

5. Tests that lock it in

Touched-blocks coverage, bidi units, east-asian rFonts round-trip, CSV import round-trip, formula-shift + csv-import units, executor/units/op-docs sync, deck-spec + layout-audit, legacy doc/ppt + html2docx tables, CLI help/registry/README/skill sync, xlsx-gateway + pptx-ops wired into CI.

6. Features (small, sharp)

• feat(sheets): infer table range from current region (fixes #298) (#371)
• feat(sheets): cross-highlight the active cell's row and column (#132)
• feat(pdf): add print range dialog (all/current/custom)
• feat(markdown): add document outline sidebar with click-to-jump
• feat(ai-search): add Tavily web search fallback via TAVILY_API_KEY
• feat(docs): per-side and inside table borders in ribbon + RTL-aware panels

Highlights (12 to start with)

• #131 fix(sheets): extend Ctrl+F beyond the loaded window on streamed workbooks
• #155 fix(slides): serialize concurrent saves to prevent file corruption
• #150 fix(ai-provider): release SSE reader when the consumer abandons the stream mid-flight
• #371 feat(sheets): infer table range from current region (fixes #298)
• #132 feat(sheets): cross-highlight the active cell's row and column
• #501 fix(project-store): reject traversal ids and clamp chat limit
• #515 fix(docx-engine): close zip handles and validate central directory
• #468 fix(docx-engine): clamp protection spinCount to prevent digest DoS
• #689 fix(cli): validate JPEG markers and image dims in image-size
• #687 fix(ai-provider): tolerate non-string media settings values
• #684 fix(pptx-render): sanitize flattenSvgPath coords and segs
• #730 fix(docs): cap AI parseTableLength at sane magnitudes
• #733 fix(docs): move hardcoded cm strings into i18n

The full list — all merged GenOffice PRs

Every PR below is merged in genspark-ai/genoffice by @aniruddhaadak80. Count in this post: 270 (complete via GitHub Search API, 719 total merged PRs of mine across 62 repos).

a11y (1)

• #498 fix(a11y): use existing i18n keys for hardcoded aria labels

agent-core (1)

• #505 fix(agent-core): guard streamText cap and extract errors

ai-panels (3)

• #192 fix(ai-panels): clear preview marker on read rejection so thumbnails retry
• #203 feat(ai-panels): RTL-aware chrome and auto-direction messages in slides/sheets/pdf/markdown (part of #13)
• #209 fix(ai-panels): clear restored history transcript on New chat (fixes #195)

ai-provider (21)

• #150 fix(ai-provider): release SSE reader when the consumer abandons the stream mid-flight
• #151 fix(ai-provider): skip malformed SSE frames instead of killing the AI turn
• #187 fix(ai-provider): sanitize Gemini tool schemas for union types and refs (fixes #186)
• #191 fix(ai-provider): return ok:false on non-JSON 200 instead of throwing
• #193 fix(ai-provider): keep default temperature for the Gemini 3 family
• #229 fix(ai-provider): allow keyless custom endpoints for local models
• #232 fix(ai-provider): trim model whitespace and remap padded retired ids
• #261 fix(ai-provider): omit temperature for OpenAI o-series reasoning models
• #303 fix(ai-provider): match JSON Content-Type case-insensitively
• #343 fix(ai-provider): match fixed-sampling models case-insensitively on every route
• #383 fix(ai-provider): bound fixed-sampling match and guard missing model
• #393 fix(ai-provider): match vision guard case-insensitively
• #415 fix(ai-provider): treat whitespace-only keys, URLs, and models as unconfigured
• #427 fix(ai-provider): flush TextDecoder at SSE stream end
• #469 fix(ai-provider): surface rate-limit and server errors in media test
• #518 fix(ai-provider): normalize watchdog timeouts
• #533 fix(ai-provider): cap streamed tool-argument buffers per tool
• #540 fix(ai-provider): cap SSE line buffer in shared stream parser
• #574 fix(ai-provider): cap streamed tool call count per turn
• #594 fix(ai-provider): validate custom and override base URLs
• #687 fix(ai-provider): tolerate non-string media settings values

ai-search (8)

• #194 feat(ai-search): add Tavily web search fallback via TAVILY_API_KEY
• #358 fix(ai-search): normalize media MIME type case-insensitively
• #426 fix(ai-search): scope copyright-host filter to the image hostname
• #465 fix(ai-search): parse gsk JSON when trailing logs follow the payload
• #503 fix(ai-search): match copyright hosts by registrable suffix
• #532 fix(ai-search): use shared host check for gsk image copyright filter
• #555 fix(ai-search): cap GSK web result count and field lengths
• #602 fix(ai-search): clamp search maxResults and query length

cli (5)

• #497 test(cli): sync help/registry/README/skill plus missing doc rows
• #512 fix(cli): harden MCP http host and file handling
• #559 fix(cli): cap safe file name length preserving extension
• #564 fix(cli): bound typo suggestion inputs and candidate scan
• #689 fix(cli): validate JPEG markers and image dims in image-size

contributing (1)

• #207 docs(contributing): document TAVILY_API_KEY in env var table

dialogs (1)

• #488 fix(dialogs): add role, focus, and escape semantics

docs (24)

• #200 feat(docs): per-side and inside table borders in ribbon (fixes #165)
• #202 feat(docs): RTL-aware AI panel chrome and auto-direction messages (part of #13)
• #223 fix(docs): complete ruler tab cycle with Bar, hide clear stops
• #224 fix(docs): drop staged attachments on New chat
• #242 fix(docs): decline Windows reserved names for auto file naming
• #257 test(docs): cover mixed-width section page boxes (issue #246)
• #258 fix(docs): preserve scroll position when zooming
• #305 fix(docs,slides): accept spaces around dashes in print ranges
• #310 fix(docs): include write-protection flags in dirty check
• #311 fix(docs): count Greek, Cyrillic, Hebrew, and Arabic words
• #321 fix(docs): detect paired bidi section tags
• #352 fix(docs): clear cross-document edit state on open, new, and save
• #353 fix(docs): count Southeast Asian scripts and full surrogate range in word count
• #412 fix(docs): exclude supplementary private-use and variation selectors from asian char count
• #494 test(docs): add bidi direction unit tests
• #511 test(docs): add touched-blocks coverage and refresh stale positions
• #550 fix(docs): validate project chat IPC payloads and bounds
• #591 fix(docs): bound page-number formatters against non-finite input
• #595 fix(docs): cap note-number repeat expansion
• #610 fix(docs): validate print-pdf-buffer page size and scale
• #660 fix(docs): guard HTML export against out-of-range hex escapes
• #662 fix(docs): cap AI length parsers at sane magnitudes
• #730 fix(docs): cap AI parseTableLength at sane magnitudes
• #733 fix(docs): move hardcoded cm strings into i18n

docx (19)

• #317 fix(docx): respect off/none/case variants in TOC bold detection
• #318 fix(docx): keep heavy leader for ptab stops
• #319 fix(docx): decode numeric char refs in notes and text patches
• #320 fix(docx): preserve edge spaces in field patches
• #348 fix(docx): split and patch on spaced and paired run-tab variants
• #355 fix(docx): read single-quoted protection attributes
• #359 fix(docx): handle numeric char refs, NBSP edges and empty self-closing runs in field patch
• #376 fix(docx): read single-quoted protection hash salt spin sid
• #381 fix(docx): read single-quoted checkbox glyph values
• #382 fix(docx): inline EQ fields with single-quoted fldCharType
• #392 fix(docx): balance single-quoted field char runs
• #402 fix(docx): honor off and case in footnote bold flags
• #403 fix(docx): honor none and case in vanish collapse
• #410 fix(docx): fold field code split across paragraphs with single-quoted fldCharType
• #421 fix(docx): read single-quoted footnote separator types and ids
• #422 fix(docx): read uppercase checkbox checked values as checked
• #423 fix(docx): read single-quoted TOC tab stop values and leaders
• #424 fix(docx): honor single-quoted xml:space preserve on part roots
• #495 test(docx): add east-asian latin rfonts round-trip test

docx-engine (11)

• #312 fix(docx-engine): read paired protection tags and on enforcement
• #468 fix(docx-engine): clamp protection spinCount to prevent digest DoS
• #509 fix(docx-engine): base-26 chart columns and hole size default
• #515 fix(docx-engine): close zip handles and validate central directory
• #556 fix(docx-engine): reject non-finite and out-of-range font sizes
• #563 fix(docx-engine): harden lazy media URL hash and part path
• #569 fix(docx-engine): sanitize non-finite ink geometry before writing
• #577 fix(docx-engine): keep overflowing measures as original bytes
• #592 fix(docx-engine): clamp EQ array column count to finite range
• #601 fix(docx-engine): bound table colSpan before grid build
• #683 fix(docx-engine): bound hostile numbering values in formatNumber

electron-utils (9)

• #240 fix(electron-utils): skip drop files whose path resolution throws
• #274 fix(electron-utils): trim and normalize URLs in safeExternalUrl
• #313 fix(electron-utils): validate fallback save dir instead of crashing later
• #396 fix(electron-utils): block trailing-dot internal hostnames
• #517 fix(electron-utils): harden renderer file resolution
• #557 fix(electron-utils): cap renderer query params and path lengths
• #562 fix(electron-utils): cap drop path resolution count and length
• #593 fix(electron-utils): normalize fetchWithSsrfGuard maxRedirects
• #612 fix(electron-utils): cap generated-image bytes and mime

file-parse (8)

• #378 fix(file-parse): keep pptx tab stops as tabs
• #400 fix(file-parse): read lowercase xlsx cell refs
• #425 fix(file-parse): include footnote and endnote text in docx extraction
• #457 fix(file-parse): extract PPTX slides in presentation order
• #464 fix(file-parse): keep empty shared-string cells empty and preserve malformed-ref text
• #508 fix(file-parse): resolve backslash and above-root pptx targets
• #530 fix(file-parse): normalize xlsx workbook rel targets
• #565 fix(file-parse): bound xlsx extraction rows, sheets, and wild columns

font-metrics (1)

• #281 fix(font-metrics): evict oldest advance-cache entry instead of clearing

general (2)

• #487 chore: wire xlsx-gateway and pptx-ops tests into CI
• #489 test: add legacy doc/ppt and html2docx tables tests

html (3)

• #398 fix(html): escape ampersands in single-quoted attrs
• #414 fix(html): escape apostrophes in single-quoted style attributes on set_style
• #571 fix(html): bound breadcrumb chain and crumb labels

html2docx (7)

• #436 fix(html2docx): map kana extensions, radicals, and astral CJK to CJK fonts
• #482 fix(html2docx): validate IR, abort image wait, restore safely
• #510 fix(html2docx): validate page background color hex
• #521 fix(html2docx): narrow shotId with type predicate (un-breaks main typecheck)
• #541 fix(html2docx): validate IR numeric geometry as finite and non-negative
• #573 fix(html2docx): sanitize render context numerics to finite geometry
• #614 fix(html2docx): guard zero-area image scale in renderer

i18n (2)

• #302 fix(i18n): require BCP-47 boundary in normalizeLang prefix match
• #394 fix(i18n): use own-property check in format

markdown (11)

• #149 fix(markdown): wait for in-flight saves during close instead of silently failing
• #222 feat(markdown): localize zoom control labels
• #260 fix(markdown): improve print quality with @page margins and high-DPI styles
• #262 test(markdown): cover print HTML builder output
• #289 feat(markdown): add document outline sidebar with click-to-jump
• #366 fix(markdown): resolve sibling images for filesystem-root files
• #390 fix(markdown): normalize pasted math delimiters to dollars
• #420 fix(markdown): serve sibling images of filesystem-root documents
• #513 fix(markdown): guard diagram rasterization and wavedrom errors
• #535 fix(markdown): drop duplicated openSource helper in roundtrip browser spec
• #549 fix(markdown): make diagram preview keyboard-operable and announced

pdf (18)

• #237 fix(pdf): suffix Windows reserved auto-rename bases
• #288 feat(pdf): add print range dialog (all/current/custom)
• #306 fix(pdf): ignore invalid font sizes when decoding text styles
• #307 fix(pdf): clamp and round channels in rgb255ToHex
• #316 fix(pdf): raise print raster to 200 DPI within a pixel budget
• #328 fix(pdf): validate colors and toggles when decoding text styles
• #344 fix(pdf): validate style-key fonts and sizes on decode and patch
• #354 fix(pdf): stream print pages with per-page budget and cleanup
• #399 fix(pdf): use length-preserving case folding in search
• #413 fix(pdf): fold AI-tool search queries the same way as the search index
• #417 fix(pdf): sanitize fields in encodeStyle so crafted styles cannot shift key fields
• #428 fix(pdf): drop non-integer page targets in printPdf
• #516 fix(pdf): harden redaction validation and layer behavior
• #520 fix(pdf): announce toasts and outline empty state
• #561 fix(pdf): cap redaction region count per request
• #576 fix(pdf): cap outline tree nesting depth
• #578 fix(pdf): validate link rects and cap overlays per page
• #600 fix(pdf): validate markup color and quads before appearance stream

pdf2docx (9)

• #304 fix(pdf2docx): merge touching intervals at minGap boundary
• #329 fix(pdf2docx): count roman-numeral TOC leader lines
• #347 fix(pdf2docx): detect underscore leaders and 4-digit leaderless pages
• #377 fix(pdf2docx): detect underscore leaders in hasDotLeaderRun
• #401 fix(pdf2docx): detect hyphen leaders in TOC entries
• #416 fix(pdf2docx): anchor arabic leader detection to line end and cap page digits
• #507 fix(pdf2docx): ignore non-finite inputs in median
• #531 fix(pdf2docx): cap empty-frame bitmap dimensions
• #615 fix(pdf2docx): require finite gap in spacing chain

pipelines (2)

• #485 test(pipelines): add deck-spec and layout-audit tests
• #609 fix(pipelines): cap outline raw JSON and image queries

pptx (7)

• #326 fix(pptx): replace solidFill when recoloring preset and system runs
• #346 fix(pptx): replace non-solid fills instead of duplicating on recolor
• #441 fix(pptx): read and strip single-quoted showMasterSp
• #443 fix(pptx): honor true/True table style flags and merges
• #446 fix(pptx): resolve and preserve single-quoted hyperlink r:id
• #447 fix(pptx): read single-quoted slide ids
• #448 fix(pptx): resolve preset colors case-insensitively

pptx-engine (6)

• #461 fix(pptx-engine): honor the 1904 date system in chart categories
• #466 fix(pptx-engine): fall back to defaults on malformed slide size
• #506 fix(pptx-engine): bound slide id allocation scan
• #567 fix(pptx-engine): land non-finite font sizes on lower bound
• #613 fix(pptx-engine): drop non-finite chart axis min/max
• #682 fix(pptx-engine): cap chartEx level sizes and idx range

pptx-ops (4)

• #484 test(pptx-ops): add executor, units, and op-docs sync tests
• #528 fix(pptx-ops): reject non-finite and negative rect dimensions in insert ops
• #538 fix(pptx-ops): reject non-finite and oversized borderWidthPt in setTableStyle
• #566 fix(pptx-ops): reject overflowing connector line widths

pptx-render (4)

• #470 fix(pptx-render): bound imageDpi cache with hashed keys
• #502 fix(pptx-render): guard viewport against degenerate sizes
• #554 fix(pptx-render): return zero for non-finite coordinate inputs
• #684 fix(pptx-render): sanitize flattenSvgPath coords and segs

print-range (1)

• #356 fix(print-range): accept em-dash, tilde and fullwidth dash variants

project-store (3)

• #501 fix(project-store): reject traversal ids and clamp chat limit
• #547 fix(project-store): clamp timeline limit and cap project names
• #575 fix(project-store): bound pending opening messages with early flush

sheets (37)

• #131 fix(sheets): extend Ctrl+F beyond the loaded window on streamed workbooks
• #132 feat(sheets): cross-highlight the active cell's row and column
• #133 fix(sheets): keep filter-hidden rows out of full-sheet find results
• #134 fix(sheets): scan the whole file for error checking on streamed workbooks
• #140 fix(sheets): row-batch over-cap sidecar reads so streaming survives big viewports
• #190 fix(sheets,slides): forward stopReason so max_tokens truncation recovers
• #214 fix(sheets): reject out-of-grid Go To refs at validation
• #215 fix(sheets): keep trailing newlines in clipboard TSV fields
• #216 fix(sheets): trim spaces in whole-cell replace paths
• #233 fix(sheets): guard replaceOccurrences against empty needle
• #239 fix(sheets): drop staged attachments on New chat
• #247 fix(sheets): label force-string popup as warning not error
• #250 test(sheets): cover multi-match-per-row find navigation (issue #220)
• #270 fix(sheets): use -1 for unfrozen axis when opening frozen panes
• #275 fix(sheets): handle escaped quotes in CSV delimiter sniffing
• #308 fix(sheets): preserve consecutive CR breaks in clipboard TSV
• #309 fix(sheets): distinguish types in aggregate distinct counts
• #322 fix(sheets): keep over-precision integers as text on CSV import
• #323 fix(sheets): carry quote state across lines when sniffing delimiter
• #324 fix(sheets): copy empty booleans as empty cells
• #325 fix(sheets): shift whole-column spans on structural edits
• #341 fix(sheets): shift whole-row spans on structural edits
• #342 fix(sheets): handle apostrophe sheet names in live formula shift
• #349 fix(sheets): preserve numbers and booleans in out-of-window Replace All
• #350 fix(sheets): clamp copy materialization to the screen extent
• #351 fix(sheets): keep single-column prose out of delimiter sniffing on import
• #360 fix(sheets): satisfy exactOptionalPropertyTypes for LazyCellMatch rawValue
• #371 feat(sheets): infer table range from current region (fixes #298)
• #379 fix(sheets): map common OOXML paper sizes instead of A4 fallback
• #418 fix(sheets): shift whole-row spans on copy/fill like whole-column spans
• #419 fix(sheets): flag modern Excel error values in Error Checking taxonomy
• #438 fix(sheets): decode BOM-less UTF-16 CSV instead of keeping NUL garbage
• #514 fix(sheets): harden MCP sheet addressing
• #548 fix(sheets): bound CSV import size before conversion
• #551 fix(sheets): validate project chat IPC payloads and bounds
• #619 fix(sheets): sanitize print-geometry numbers before emitting print HTML
• #671 fix(sheets): guard defaultThreshold against zero count

shell (12)

• #156 fix(shell): surface project IPC failures and AI settings save/test errors
• #221 fix(shell): harden Home rename validation
• #230 fix(shell): reject Windows reserved and trailing-dot rename names
• #231 fix(shell): normalize recents ext filter dots case and whitespace
• #273 fix(shell): include xls in the sheets recents filter family
• #357 fix(shell): share ext family between recents and starred filters
• #397 fix(shell): reject trailing spaces in rename gate
• #411 fix(shell): validate raw rename name before trimming so trailing-space gate holds
• #529 fix(shell): route tab lookups through canonical path comparison
• #537 fix(shell): validate tab activate and close IPC ids
• #560 fix(shell): validate home openPath IPC string and length
• #622 fix(shell): cap CLI-runner output buffering

slides (21)

• #141 fix(slides): sweep generated-page temp files with a TTL at startup
• #155 fix(slides): serialize concurrent saves to prevent file corruption
• #160 fix(slides): skip width overflow check on vertical text bodies
• #218 fix(slides): route saveAs through the save queue
• #219 fix(slides): clear answered clarifications on New chat
• #234 fix(slides): clear staged attachments on New chat
• #287 fix(slides): respect AI composer spellcheck pref
• #327 fix(slides): escape image sources in print HTML
• #345 fix(slides): normalize corrupt print and export page geometry
• #380 fix(slides): use theme tokens in cutout dialog
• #384 fix(slides): localize find replace close button label
• #395 fix(slides): reject non-RRGGBB input in recent colors
• #439 fix(slides): cap and dedupe recent colors on read
• #534 fix(slides): reset save state on stale current-slide index
• #552 fix(slides): validate project chat IPC payloads and bounds
• #570 fix(slides): guard print dialog against stale current-slide index
• #599 fix(slides): validate ink descr payload before decode
• #623 fix(slides): cap metafile raster input bytes in image loader
• #668 fix(slides): reject out-of-range start slide in play order
• #672 fix(slides): cap pasted-image IPC bytes at 5MB
• #673 fix(slides): correct Custom Show counts and dialog semantics

tests (1)

• #314 fix(tests): repair two red-main test expectations

xlsx-gateway (7)

• #467 fix(xlsx-gateway): harden cell address parsing and cap range expansion
• #483 test(xlsx-gateway): add formula-shift and csv-import unit tests
• #496 test(xlsx-gateway): add csv import round-trip tests
• #504 fix(xlsx-gateway): reject impossible DV dates and times
• #519 fix(xlsx-gateway): validate style color hex input
• #553 fix(xlsx-gateway): cap CSV row and column counts during parse
• #598 fix(xlsx-gateway): order non-finite keys deterministically in sort

What I learned

1. Real-world files don't follow the spec. If Word can emit it, you must parse it: single quotes, uppercase booleans, paired tags, backslashes.
2. Clamp at the boundary, not in the middle. Validate IPC, IDs, sizes, counts, and numerics where they enter — renderers stay simple.
3. Non-finite is a vulnerability class. NaN/Infinity handling deserves the same discipline as XSS/traversal.
4. Tiny PRs merge faster. One fix + one test beats a 2,000-line refactor.
5. A11y/i18n is correctness. If a screen reader or RTL user can't use it, it's broken.

What's next

Continuing the hardening pass (latest: #730 AI parseTableLength caps, #733 cm→i18n, #689 CLI JPEG validation, #687 media-settings tolerance, #684 SVG flattening). If you maintain an Office parser, start with hostile-input fuzzing around numbers, quotes, and encodings — you'll find bugs in an afternoon.

---

All PRs merged in https://github.com/genspark-ai/genoffice — author: https://github.com/aniruddhaadak80