feat(M4): FPP-forward farbling + dark mode unlock - #20
Merged
Merged
Conversation
Switch anti-fingerprinting from RFP (uniform, "stands out", per-read canvas randomization) to FPP (per-origin deterministic farbling), the roadmap's "convincing common fingerprint" posture. The three FP modes are mutually exclusive, so RFP must be off for FPP to run. assets/neonwolf.overrides.cfg: - privacy.resistFingerprinting=false, privacy.fingerprintingProtection=true, remoteOverrides.enabled=false. - fingerprintingProtection.overrides adds WebGL randomization, hw-concurrency pin, audio/font/timer protection on top of FPP's default canvas farbling. UA/screen uniform spoofs deliberately omitted (real-for-platform is more common); timezone left real (its JS override is RFP-pref-gated, and a timezone/IP mismatch is itself a tell). - webgl.disabled=false so WebGL is farbled (WebGL-off is a conspicuous tell). - layout.css.prefers-color-scheme.content-override=2 (auto). Verified locally (Marionette): canvas farbling deterministic per-origin (Brave-style), hardwareConcurrency pinned to 4, and DARK MODE unlocked (prefers-color-scheme honors the override under FPP; RFP was forcing light). NOT merged: per plan, ships only behind a full browserleaks/coveryourtracks + WebGL pass (needs a real GL context + network — headful), proving it's more convincing, not less protected. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
AI cowork.md (the Strong/Weak-AI playbook), GROK.md (the handoff artifact it prescribes), and docs/agents/ are local working scratch, never repo content. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Empirical validation of the FPP posture found WebGL still dead in content: LibreWolf's webgl-permission.patch gates context creation behind a per-site 'webgl' permission (librewolf.webgl.prompt, compiled default TRUE) and ships the doorhanger hidden — silent deny everywhere, the exact same conspicuous tell webgl.disabled=false was meant to remove. Turn the gate off; WebGL now runs farbled (+WebGLRandomization per-eTLD+1 readback noise, verified deterministic per-origin/per-session; +WebGLRenderInfo spoofs Mozilla/Mozilla). Full headful validation (Marionette on Wayland, real network): canvas + WebGL farble per-eTLD+1, stable across reloads, re-keyed per session, plain control identical across domains with all tiers off; hwConcurrency pinned 4; timezone/screen/UA real; dark mode honored; CoverYourTracks verdict 'strong protection against Web tracking'. Known gap: audio readback is NOT farbled (FPP has no audio randomization target — RFP-only protection; follow-up). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…scriptions UI Shields panel (M4 'integrate with Shields'): - fingerprinting toggle now drives privacy.fingerprintingProtection (the M4 posture pref), label 'Farble fingerprinting' - new per-site 'Farble on this site' subrow writing an -AllTargets entry to BOTH privacy.fingerprintingProtection.granularOverrides AND the baseline tier's granularOverrides (canvas farbling rides baseline FPP — verified empirically); foreign/webcompat entries are never touched, unreadable pref hides the row (all-or-nothing writes), PSL-hosted eTLD+1 edge documented Element picker (M5): real selector generator replacing the placeholder — unique #id fast path, bottom-up max-4-segment path with stable-class filtering (hashed CSS-module classes rejected), :nth-of-type disambiguation, document-uniqueness testing, simpleSelector fallback; pure DOM reads only. Lists page (M5): 'Subscribed lists' manager bound to neonwolf.shields.lists.subscriptions — https-only validation, dedupe, live pref observer, Remove per row, synthwave-styled via shared filters classes. Implemented by Grok (executor) from Claude's spec; line-by-line reviewed, staged into the built tree and Marionette-verified end-to-end (granular-pref roundtrip incl. corrupt-pref no-clobber, page render screenshot). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
UBONetFilter now compiles neonwolf.shields.lists.subscriptions (newline- separated https:// URLs, managed by the Shields lists/filters pages) as extra engine lists on every live refresh, and a pref observer triggers an immediate refresh on subscribe/unsubscribe. Fetch-only by design: no bundled fallback (a failed fetch drops that subscription until the next cycle), 1-byte floor (personal lists can be tiny) but a 20 MB cap so a hostile URL can't balloon the compile; invalid/non-https lines are ignored. No selfie write — like the bundled-list refresh, subscriptions apply in-memory after the first refresh of a session. Marionette-verified end-to-end against the built tree: subscribe to easylistgermany -> refresh fires -> adnx.de script blocked; unsubscribe -> refresh -> unblocked again. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Owner
Author
|
M4 merge gate: PASSED (overnight validation run, headful Marionette on Wayland, real network, built tree)
Known gap for a follow-up: audio readback is not farbled (FPP has no audio noise target; that protection was RFP-only). 🤖 Generated with Claude Code |
…k gap First live adblock benchmark (adblock.turtlecute.org, the maintained d3ward replacement — d3ward is archived) scored 116/133. Engine-side probing of every test host pinned the shortfall to 9 tracker/ad network hosts unmatched by all ten bundled lists: TikTok ads/events APIs, Yahoo ad-tech + DMP, and the Yandex Metrika/AppMetrica apex (bundled lists only covered paths/subdomains). Add a curated assets/adblock/neonwolf-extra.txt (bundled-only, no fetch URL — served from the dump) with those 9 hosts as $third-party rules so first-party use of each console still works. Wired into gen-adblock-dump.py's LISTS, the list_names pref, and the lists-page display. Re-benchmarked against the rebuilt engine: 125/133 = 94%, all 131 network hosts now blocked (verified via direct matchRequest). The remaining 8 are cosmetic-filter + ad-script-loading checks — the known M2 scriptlet gap (scriptlets pref-disabled pending the document_start race fix), not a list gap. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Replace the archived d3ward metric with adblock.turtlecute.org throughout; record M4 (FPP-forward farbling, validated PR #20) and M5 (picker, custom filters, subscriptions) as functionally complete with 2026-07-08 progress notes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Owner
Author
|
Follow-up shipped: adblock benchmark gap closed → 94%. d3ward is archived; benchmarked on adblock.turtlecute.org (the maintained d3ward-style replacement). First run was 116/133 (87%). Engine-side probing pinned the shortfall to 9 tracker/ad hosts unmatched by all ten bundled lists (TikTok ads/events APIs, Yahoo ad-tech + DMP, Yandex Metrika/AppMetrica apex). Added a curated bundled list Also refreshed PLAN_OF_ACTION.md (metric d3ward→turtlecute; M4/M5 marked shipped). 🤖 Generated with Claude Code |
Complete the Google Safe Browsing opt-out for this privacy-focused build. The LibreWolf baseline already disables the SB malware/phishing/blockedURIs/downloads features and blanks the google4 + legacy-google gethash/update URLs, but leaves the newer google5 provider (stock FF152 ships it enabled) and the google4 dataSharing toggle untouched — a residual safebrowsing.googleapis.com phone-home. Turn those off too so no Google SB lookup remains; malware/phishing coverage comes from the bundled uBO lists (ubo-badware et al.) via the native engine. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The notifier queried GitHub /releases/latest, which excludes prereleases and drafts. Beta builds ship as GitHub *prereleases*, so a beta tester would never be told a newer beta exists. Query /releases?per_page=30 instead and scan for the highest-version non-draft release (prereleases eligible, drafts skipped), reusing the existing _isNewer comparator and keeping the throttle, enable gate, and error handling unchanged. Verified against the live GitHub API: scans 5 releases, correctly picks 152.0.1-3, treats the 152.0.1-1 prerelease as eligible. Executor: Grok (worktree, no-commit); reviewed, tested, and captured by Claude. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ection Public-facing beta docs for Linux + Windows testers: download links, how to install the unsigned builds past SmartScreen / run the AppImage, how the notify-only update check works (no auto-update by design), and a Known Issues list (adblock ~94% pending scriptlets, audio not farbled, no macOS build, unsigned, no built-in password manager, Windows less-tested). README gains a short Beta downloads section linking it. Executor: Grok (worktree, no-commit, facts-only spec); reviewed line-by-line against the known facts and captured by Claude. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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
1 participant
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.
Switches anti-fingerprinting from RFP (uniform, detectable, per-read canvas randomization) to FPP (per-origin deterministic farbling) — the roadmap's 'convincing common fingerprint' posture. The FP modes are mutually exclusive, so RFP goes off.
Also unlocks dark mode (a stated adoption blocker): RFP was forcing
prefers-color-scheme: light; under FPP the native 'Website appearance' setting works.Verified locally (Marionette, no rebuild — runtime/startup prefs)
Merge gate (NOT done yet — needs headful GL + network)
browserleaks canvas/webgl/audio/fonts (site-varying, protected) + coveryourtracks (randomized, no privacy-browser tell). WebGL can't be tested headless (no GL context).
🤖 Generated with Claude Code