abr00kx/4chan-next-audit

4chan-neXT Comprehensive Security Audit & Codebase Provenance

A forensic open-source intelligence (OSINT) and static security audit of the 4chan-neXT userscript project and its landing domain https://4chan-next.com/.

---

1. Executive Summary & Domain Audit

• Domain Name: 4chan-next.com
• Registrar: NameCheap, Inc. (IANA ID: 1068)
• Registration Date: 2026-05-30T03:09:24Z
• Expiration Date: 2027-05-30T03:09:24Z
• WHOIS Status: Masked via Withheld for Privacy LLC, Lewes, Delaware, US.
• Hosting Infrastructure: GitHub Pages via Fastly Anycast CDN (185.199.108.153 - 185.199.111.153), HTTP header server: GitHub.com.
• Associated Web Repositories:
  • Landing Website Source: cercos/next-webpage (hosts static index.html, styles.css, and CNAME: 4chan-next.com).
  • Software Application Source: cercos/4chan-neXT (the active userscript / Chrome extension fork).
• Project Maintainer: Derek (cercos, commit email dcostroff@gmail.com).

---

2. Codebase Bloat & Developer Sloppiness

A major concern identified during initial repository triage was the massive commit diffs (+147,223 / -61,615 lines across 251 files) in the neXT fork era. A deeper inspection of git history reveals significant development sloppiness regarding artifact tracking:

The builds/ Bundle Bloat

Rather than treating compiled single-file bundles as release assets, the maintainers continually committed monolithic production bundles directly into git:

• builds/4chan-neXT.user.js contains 47,347 lines of compiled code.
• On day one (commit 810e55db3), Derek replaced the prior 4chan-XT.user.js (-26,447 lines) with 4chan-neXT.user.js (+47,347 lines), bloating the git log with a ~73,000-line single-commit diff.
• Across the 222 commits in this era, the builds/ directory generated 149,361 lines of sheer git noise across just 16 files as minified scripts and CRX assets were rebuilt and committed repeatedly.
• Additionally, automated TypeScript conversion scripts (ratcheting noImplicitAny, strict null checks, and noImplicitThis across commits 8e34a9dd0 through 1d12b47a6) modified 70–100 files simultaneously without isolating structural changes from generated bundles.

---

3. Security Findings & Auditability

Despite the extreme commit bloat, the repository is safe and fully auditable.

Isolating the generated builds/ directory from human-written code in src/ allows for high-confidence forensic analysis:

1. Deterministic & Reproducible Builds: Compiling from source (npm run build:userscript) on a clean checkout generates an artifact that matches builds/4chan-neXT.user.js byte-for-byte with zero git diff. No hidden backdoors or unauthorized scripts are injected into distribution files.
2. Compiler & Test Integrity:
   • TypeScript typecheck (tsc --noEmit -p tsconfig.typecheck.json && tsc --noEmit -p tsconfig.tests.json) passes cleanly with 0 errors.
   • Vitest test suite (npm test) passes 284 out of 284 unit tests with 100% pass rate.
3. Static Security Vectors:
   • Dynamic code execution (eval(), new Function()): 0 occurrences.
   • Exfiltration APIs (sendBeacon(), WebSocket, EventSource, GM_download()): 0 occurrences.
   • Credential interception (document.cookie capture, form harvesting): 0 occurrences.
   • Network destinations: Strictly bounded to 4chan APIs/CDNs, community imageboard archives (4chenz.github.io, nstepien.github.io), and standard media embed providers.
   • Dōgiri Editor (src/Posting/Dogiri*.ts): 5,095 lines of WebCodecs video processing that operates 100% client-side offline with zero external network connectivity.

---

4. Contributor Audit Breakdown

Across all 222 commits in the neXT era (810e55db3..HEAD):

• Johnny (saxamaphone69, 877480+saxamaphone69@users.noreply.github.com) — 3 Commits:
  • Clean: purely cosmetic adjustments (fourchan-next document class name, button title fixes).
• Kody (mkody, im@kdy.ch) — 1 Commit:
  • Clean: updated tweet embed endpoint from dead twitframe.com to community proxy tf.rita.moe.
• Derek (cercos, dcostroff@gmail.com) — 218 Commits:
  • Clean: legitimate software maintenance, typing migration, UI enhancements, and local video toolchain additions.

---

5. Root Cause Analysis: The HoloGOD Tripcode Filter Bug

On 4chan /g/ (Thread 110021812, Post 110031353/110031361), users noted:

"kinda unrelated, but anyone know what's up with tripcodes on 4chan-neXT? sometimes they get merged into the name field, other times not. kinda strange" — Anonymous, No.110031353

And reported that the script's filtering function fails to filter user HoloGOD (She/Her) !!zcleypcJXz2.

Technical Root Cause

1. 4chan Microdata Markup Shift: In server-rendered desktop HTML, 4chan updated post markup to include schema.org/Person metadata:

   <div class="postInfo desktop" id="pi110031305">
     <span class="nameBlock">
       <span class="name" itemprop="author" itemscope itemtype="https://schema.org/Person">
         <span itemprop="name">HoloGOD (She/Her) !!zcleypcJXz2</span>
       </span>
     </span>
   </div>
   

   Crucially, 4chan's desktop backend omits the <span class="postertrip"> tag completely, merging the tripcode directly into the .name span.
2. DOM Parser Assumptions in Post.ts: 4chan-neXT's DOM parser (src/classes/Post.ts / src/site/SW.yotsuba.tsx) expects distinct selectors:

   name: '.name',
   tripcode: '.postertrip'
   

   On initial page load, post.nodes.tripcode evaluates to undefined, and post.nodes.name evaluates to "HoloGOD (She/Her) !!zcleypcJXz2".
3. The "Sometimes Merged, Sometimes Not" Paradox:
   • Initial Desktop Page Load: Native HTML contains no .postertrip; name and tripcode are merged. Tripcode filters fail; exact name filters fail.
   • Thread Updater (JSON API): PostInfoHtml.tsx builds elements from JSON (data.name and data.trip), generating separate .name and .postertrip elements.
   • Consequently, posts appearing dynamically via the thread updater match filters correctly, while initial-load posts fail to match, causing severe visual and filtering inconsistency across browser tabs and reloads.

The Resolution

In src/site/SW.yotsuba.tsx (SWYotsuba.parseNodes):

• Detects when !nodes.tripcode and extracts the tripcode from either the mobile markup (.postInfoM.mobile .postertrip) or via tripcode syntax pattern (/\s+(!{1,2}[\w./+]+)$/).
• Uses String.prototype.endsWith and slice to safely strip the tripcode suffix from nodes.name without regular expression escaping pitfalls.
• Reconstructs the <span class="postertrip"> element and attaches it after nodes.name in desktop postInfo.
• Re-enables 100% reliable filter matching and post highlighting across both initial page loads and thread updates.
• Unit tested in src/site/SW.yotsuba.test.ts (3/3 test cases passing; 287/287 total test suite passing).

---

6. License

This audit report and derivative works are licensed under the GNU Affero General Public License v3.0 (AGPLv3).