* feat(ts): import the TypeScript launcher port from feat/captchakrakenAndJSSupport CAPTCHA support is left out; this branch is the JS/TS driver only. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(ts): drop the CAPTCHA wiring left behind by the import Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(ts): port fpgen to TypeScript, on the same pinned model fpgen is not on npm. The port reads scripts/data/fpgen-model.json and checks its sha256 with TLS on, never fpgen's own first-release download. Everything that does not depend on the random draw is identical to Python (network, value lookups, trace probabilities, conditions, errors); the draws are held to Python's distributions by chi-square tests. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(ts): identity layer at parity with pythonlib A bit-exact port of CPython's random.Random, numpy's PCG64 choice and orjson's serialisation, so identity_salt/identity_seed and every seeded draw (fonts, voices, media devices, WebGL, noise seeds) come out identical to Python for the same identity. coherence.py, presets and screen/window fixes are ported, and golden fixtures recorded from pythonlib hold all of it to exact equality. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(ts): launcher at parity with pythonlib's launch_options launch_options() now produces pythonlib's output byte for byte (CAMOU_CONFIG, CAMOU_PREFS_N, prefs, env, fontconfig, warnings) over 89 recorded scenarios. Ports core pinning, geolocation, locales, fontprobe, the async API, and the pkgman/multiversion integrity checks. An opt-in e2e suite launches a real build through both launchers and compares what a page sees. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci: test the TypeScript package, and publish it to npm like pypi - ci/run_typescript.py writes the `typescript` gate (typecheck, lint, vitest with the pythonlib golden tests) and, with --browser, `typescript_browser` (the e2e suite against the browser under test). Both are required by the gate. - publish-npm.yml mirrors publish-pypi.yml: workflow_dispatch, checks, build, scripts/check-pack.mjs (version == pythonlib, every data file shipped, the tarball installs and imports), then publish via npm trusted publishing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): e2e that holds on any browser, NewContext, and the package on every PR - Parity (TS == Python on the same binary) stays strict everywhere; whether the browser honours the config is asserted only on a binary whose properties.json knows every key the launcher sets, and otherwise skips naming the missing keys. A driver-only pull request is tested against the published release, which lags the launcher (beta.30 predates #779), so this is what makes the suite meaningful there instead of red on skew it cannot fix. - New: NewContext in a real browser -- a per-context identity that differs from the launch identity and from a sibling context, and equals Python's. - python_probe.py keeps stdout for its JSON (pythonlib prints "Skipping unknown patch" there), and a non-JSON reply now fails fast instead of hanging 240 s. - The typescript gate builds the package and runs scripts/check-pack.mjs, so a packaging mistake fails the pull request that makes it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ci): commit the launch fixtures, and fetch the browser from its real directory - The root .gitignore ignores every path named `launch` (local build output), which silently dropped typescript/tests/fixtures/launch/ -- the launch_options() goldens -- from the branch. Re-included in typescript/.gitignore. - fetch-browser read camoufox-bin from `camoufox path`, the cache ROOT, but multiversion installs each build under browsers/<channel>/<version>/, so the job has failed on every driver-only pull request since #772. It now resolves the active build as the launcher does (pkgman.camoufox_path). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci(ts): give the typescript job pythonlib, so the cross-language checks run Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): fpgen model install is safe across processes Processes installing into an empty cache at once each downloaded the model, and one's install deleted the values.dat another had just decompressed, which then failed its next lookup with ENOENT. Seen with vitest's parallel files on a cold cache; a worker pool on a fresh machine would hit it too. - ensureModel() installs under a cross-process lock (an atomic mkdir, stale after 10 min) and re-checks what is installed once it holds it. - values.dat is only removed when the model is actually being replaced. - The model keeps values.dat open, instead of reopening it on every lookup. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test: make local runs and CI see the same test suite Three ways the suite passed here and not on the runner, each fixed at its cause: - The root .gitignore's bare `launch` rule (for the Go launcher binary) ignored every path segment named launch, so tests/fixtures/launch/ never reached git. Anchored to /launch; and ci/run_typescript.py now fails when any file under typescript/{src,tests,scripts} is git-ignored, which would have caught it on the machine that wrote the fixtures. - A missing prerequisite (fpgen model, pythonlib venv, fontTools, Xvfb, a font directory) skipped its tests, and a skip reads as green. tests/prereq.ts now fails them under CI unless the job names the gap in CAMOUFOX_TEST_ALLOW_MISSING. The typescript job installs all of them. The font-name check read one developer's local browser bundle; it now reads /usr/share/fonts (or CAMOUFOX_TEST_FONT_DIR), and CI installs a .ttc set. - The fpgen install race surfaced only on a cold cache, by accident. It now has deterministic tests: a same-model reinstall keeps values.dat (verified to fail on the old code), the lock admits one holder and releases on error, and a stale lock is reclaimed. Also: the browser gate runs only the e2e file, and the e2e probe and the virtual-display test time-box each await, so a hang names its step instead of reporting a bare 240 s timeout. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): hold headless="virtual" to Python's, not to headless On the runner (no media hardware), published beta.31 never settles enumerateDevices() in a headful window while headless answers -- the named timeout in the probe caught it. That is a browser property, so like the other page-vs-config checks it moves to a test that runs on a binary current with the launcher; the virtual-display test now requires the same page as Python's headless="virtual" on the same binary. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(build-tester): accept 18 and 22 cores, as real hardware reports plausibleHWC's list of common core counts lacked 18 and 22 -- Intel Meteor Lake laptops (Core Ultra 5 125H, Core Ultra 7 155H), and 22 is in 8 recorded presets. build-tester draws random presets, so a run that picked one of the two Linux presets reporting 22 failed: about one run in eleven, on any pull request. A CI self-test now fails if the list rejects any core count pythonlib can present (the presets and PLAUSIBLE_CORE_COUNTS). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ci): test on the published release only when it matches this tree A pull request that did not touch the browser was always tested against the published release. The patch guards and suites come from the checkout, so once a browser change was merged but not yet released (#779, on top of beta.31), every driver-only pull request ran #779's guards against a browser without #779 -- eight guards failed on #785, which changes no browser source. resolve now also compares the tree's browser sources with the tag the release was cut from (v<version>-<release> from upstream.sh), and builds when they differ or the tag does not exist. Building restores the base branch's cached browser when its compiled half matches -- main's #779 build, here -- so the extra cost is a cache restore, not a compile. Self-tests run the workflow's own scope step in a scratch repo for the four cases. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): say when CI's browser cannot present the configured locale With #785 finally tested on a browser current with the tree (main's cached #779 build), every TS-vs-Python parity check passed and the page-vs-config check failed: a de-DE/fr-FR identity presented en-US. Python presents the same on that binary. CI tests the build job's unpackaged dist/bin, whose res/multilocale.txt lists en-US only -- scripts/package.py injects the langpacks, and CI never packages. So no CI suite had ever run a non-English locale on a browser that has one. The e2e locale assertions now run when the binary under test packages the configured locale (read from res/multilocale.txt, loose or in omni.ja), and otherwise go through prerequisite("packaged-locales"), which fails in CI unless the job names the gap. The typescript (browser) job names it, with the reason; the rest of the page-vs-config check stays strict. On a packaged #779 build all of it, locale included, passes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(guards): judge the query-cost probes on a median, not one sample stock-parity-probes timed each getter once. On a shared runner one GC pause or CPU-steal spike decided the verdict: navigator.hardwareConcurrency took 77 ms against a 50 ms allowance on the same restored build that passed the run before. Each pair is now timed five times, interleaved, and compared by median. The regressions these catch (a sync IPC per read, ~240 ms over the loop) cost extra on every read, so they move the median; verified by giving the getter a constant ~4 us of extra work per read -- 86 ms median, FAIL -- while the healthy build passes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): give the headful e2e page focus before probing it enumerateDevices() intermittently never settled in the headless="virtual" test on CI (passed one run, timed out the next, same build). Firefox defers device enumeration until the document has focus -- LEAKS row 57 recorded the same for a background tab -- and headless mode fakes focus while a headful window on a bare Xvfb, with no window manager, only sometimes receives it. A user's window has focus, so both launchers' virtual-display probes now bring the page to the front first. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore: remove build tooling nothing uses - The developer UI (scripts/developer.py, `make edits`). It depended on easygui, which no requirements file declares, and every action it offered is a Makefile target: patch, unpatch, workspace, revert, diff. Its two helpers in scripts/_mixin.py (is_bootstrap_patch, patch) had no other callers. - legacy/, the Go launcher deprecated in 2024-11. Nothing built or shipped it. Its Makefile targets and scripts/run-pw.py go with it, and so does Go from every dependency list and workflow. - jsonvv/ and settings/camoucfg.jvv. Nothing read the .jvv schema: config is validated against settings/properties.json, and the two had already drifted. The jsonvv package stays on PyPI. - Scripts with no caller: bootstrap.py, moztree, setup-wasi-linux.sh, package-helper.sh, install-local-build.sh, mozfetch.sh (copied into lw/ but never packaged), examples/. - The pre-ESM Juggler copies JugglerFrameParent.jsm and JugglerFrameChild.jsm, and hidden-scrollbars.css. Juggler loads the .sys.mjs actors and deliberately no stylesheet, but jar.mn still packaged all three. - patches/librewolf/*.opt, which list_patches() never picks up; the roverfox second pass in patch.py, whose directory no longer exists; the unread --no-settings-pane option. - The CAMOUFOX_PASSWD secret passed to `make fetch` and closedsrc_rev in upstream.sh, which nothing reads. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(python): remove dead helpers and a stale dependency None of these had a caller: - pkgman: is_supported_path, extract_zip, cleanup and set_version, left over from the single-directory install. cleanup() would have deleted every installed browser version. - multiversion.get_cached_repo_names, CONSTRAINTS.as_range, fingerprints._load_os_voices, utils._clean_locals, and unused imports. Also: - The "Apify Fingerprints" row in `camoufox version`, which has read "?" since fpgen replaced BrowserForge. - lxml is no longer a dependency; nothing imports it. - The geoip extra now names maxminddb, the module geolocation.py actually imports, rather than getting it transitively through geoip2. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs: show the cursor paths humanize=True actually produces The README's cursor video showed the Bezier generator Camoufox replaced with Cursory's recorded trajectories. scripts/cursor-demo.py drives a real build with humanize=True and records every mousemove event the page receives. It writes assets/humanize-cursor.svg, an animated replay at the recorded speed, so what the figure shows is what a site sees. The script cannot change the binary, so ci/browser_inputs.py lists it as non-native and editing it does not invalidate the cached browser. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(python): stop naming BrowserForge in user-facing text fpgen replaced BrowserForge, but two LeakWarnings, the NonFirefoxFingerprint message and the fingerprint_preset docstring still named it. One warning also linked to a README anchor that no longer exists. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs: one AGENTS.md for every agent, a roadmap, and docs that match the code - AGENTS.md holds the engineering rules for any coding agent, plus the repo map, build, patch and test commands that CLAUDE.md used to carry. CLAUDE.md now only imports it, so there is one set of rules. ci/tribal-rules.yml is the record of settled decisions it points to. - ROADMAP.md lists planned work, each item linked to its issue. - README: - fpgen and the coherence check replace BrowserForge; - the patch workflow uses the make targets instead of the removed developer UI; - letter-spacing noise is described as off by default, as it is. - docs/: - beta-testing-ff146.md removed; - patch-upgrading-guide rewritten around the make targets; - per-context-patches without the canvas patch that no longer exists, and with measured preset counts; - playwright-maintenance without the JSM wrapper that does not exist; - smaller fixes in MEDIA-DEVICES, input-dispatch and FONTS. - ci/README: every job, and the real shard, skiplist and entry-point lists. - pythonlib, tester and patch-dependency READMEs corrected against the code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(pythonlib): handle headless='virtual' in launch_server launch_server() is documented to take the same arguments as Camoufox(), but passed headless='virtual' straight to launch_options(), so the server launched with no Xvfb display. Start a VirtualDisplay the way Camoufox() does, launch headful on it, and kill it when the server process exits or the launch fails. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(python): remove fontprobe, which nothing called fontprobe listed the fonts installed on the host, for a `camoufox fonts` command that was never added. It has nothing to do with the font bundle Camoufox serves to pages. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(license): the Python launcher is MIT; the browser stays MPL-2.0 The Python package has always been published to PyPI as MIT (#727), but pythonlib/ shipped no licence file, and the repo's LICENSE is the browser's MPL-2.0. MPL is copyleft per file. It covers the modified Firefox sources, not a separate launcher that drives the browser over Playwright. So pythonlib/LICENSE now carries the MIT text its metadata already declares, and a Licensing section in the README says which part is which. Closes #727. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): fingerprint_preset=False no longer turns presets on launch_options checked `fingerprint_preset is not None`, so passing False drew a random bundled preset, the opposite of what was asked. It now uses a truthiness check, and a test proves that None and False never draw a preset. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci: bring the release workflow in line with the tests build job build.yml had drifted from tests.yml. It ran actions at v1/v2 on a retired Node runtime, prepared the source tree with bare make calls that fail the whole release on one dropped connection, and built with a different Python than every pull request is tested with. - Pin every action by commit SHA, at the major versions tests.yml uses (checkout v4, setup-python v5, upload/download-artifact v4, the same remove-unwanted-software SHA), and action-gh-release v2. The release job holds contents: write, so it should not follow a movable tag. - Prepare the tree with `python3 -m ci.run_prepare`, as the tests build job does. BUILD_TARGET is set from the matrix so `make dir` writes the right mozconfig and Rust targets; multibuild.py then finds _READY and builds without re-patching. mach's toolchain bootstrap ignores the mozconfig, so running it after `dir` bootstraps the same toolchains. - Build with Python 3.12, the version the tests build job compiles with. - Default the workflow to no permissions; the build job gets contents: read and the release job keeps contents: write. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): NewContext looks up a proxy's exit IP through the right URL, or fails NewContext derives the context's WebRTC IP and timezone from the proxy's exit IP. That lookup had two defects, and both left the context showing the host's values while its traffic went through the proxy: - It built its own proxy URL with urlparse, which reads a scheme-less server such as "1.2.3.4:8080" (a form Playwright accepts) as scheme "1.2.3.4" with no host. urllib could not use a SOCKS proxy at all. - Any failure was swallowed, and the context opened without the values. The URL is now built with Proxy.as_string(), which the geoip launch path already uses (scheme-less means http). The lookup goes through requests, which handles SOCKS, and a failed lookup raises InvalidIP, naming the two options that skip it. The tests cover scheme-less, http and socks5 servers with credentials, both failure modes, and the case where no lookup is needed, for NewContext and AsyncNewContext. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix: stop generating a canvas seed, and drop config keys nothing reads The browser has not noised the canvas since #528, and no patch reads canvas:seed (#721). The launcher still drew one on every launch and sent it through CAMOU_CONFIG, and NewContext called a setCanvasSeed that does not exist. They no longer do. For users this changes nothing on any browser since #528: the value was ignored. A config that still passes canvas:seed gets the usual "Skipping unknown patch" notice instead of silence. On a browser from before #528, the launcher no longer turns canvas noise on, which is the behaviour #528 chose. The same audit found more keys declared in settings/properties.json that no patch or Juggler file reads, so setting them did nothing: - canvas:aaOffset, canvas:aaCapOffset - memorysaver, pdfViewerEnabled, webrtc:localipv4/6 - navigator.onLine, navigator.cookieEnabled, navigator.languages - navigator.appCodeName, appName, product, productSub. Firefox reports these constants itself, so fpgen.yml no longer maps them. - webGl:parameters:blockIfNotDefined and its WebGL2 twin test_config_schema now checks this direction too: every declared key must be read by the browser, unless it is listed with a reason. Three are listed: locale:script and navigator.doNotTrack, which the launcher applies itself, and navigator.buildID (#780). The build-tester grading followed the same wrong premise. It tracked canvas collisions as an unfixed per-context leak. A canvas that is rendered rather than noised follows the fonts and GPU, as it does on real machines, so canvas collisions are now counted with the other device-level values. The tribal rule that recorded it as an open question is now a settled one, canvas-is-not-noised, with an automated check. Closes #721. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(ts): remove dead helpers None of these had a caller: - pkgman: isSupportedPath, extractZip, cleanup and setVersion, left over from the single-directory install. cleanup() would have deleted every installed browser version. - multiversion getCachedRepoNames and getCachedVersions, CONSTRAINTS.asRange, removeMmdb (Python keeps its twins for the GUI) and pycompat pySorted. - The "Apify Fingerprints" row in `camoufox version`, which read "?". utils.ts now calls noiseSeedsFromIdentity instead of repeating its two formulas inline, so the tested function is the one that runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(ts): remove fontprobe, which nothing called fontprobe.ts listed the fonts installed on the host, for a `camoufox fonts` command neither launcher has. It has nothing to do with the font bundle Camoufox serves to pages. Its parity test goes with it, and so do the CI prerequisites only that test needed: fonttools and the extra font packages. (The Python twin is removed in #787.) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(ts): license the launcher MIT, with third-party notices The TypeScript launcher is a port of pythonlib, which has always been published to PyPI as MIT (#727); the MPL-2.0 of the browser covers the modified Firefox sources, not a launcher that drives it over Playwright. THIRD_PARTY_NOTICES.md ships in the npm package with the notices for the code the port translates: fpgen (Apache-2.0), CPython's random (the MT19937 BSD notice and the PSF licence), and NumPy's SeedSequence and PCG64 (BSD-3 and MIT). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs: document the TypeScript package outside typescript/ The README, CONTRIBUTING, ci/README and the issue templates did not mention the npm package or its two CI gates. ci/README also still said driver-only pull requests never build. Since the scope step started comparing browser sources against the release tag, they build whenever the release is behind. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): repair devicePixelRatio the same way on every launch The DPR repair snaps an off-grid ratio to the nearest real scaling step and keeps the first of two equally near steps. The steps were frozenset literals, and a frozenset literal iterates in one order when the module is compiled from source and another when it is loaded back from a .pyc. So a midpoint such as 1.125 became 1.25 on the first launch after an install and 1 on every launch after it: the same pinned identity presented two different devicePixelRatio values. The steps are now ascending tuples, so a tie always goes to the lower step. The test runs the repair in two fresh interpreters that share a bytecode cache, compiling in the first and loading in the second. It failed before this change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): mirror #787's pythonlib fixes The TypeScript side of the behaviour #787 changes in pythonlib, so the port stays at parity: - fingerprint_preset=false no longer draws a preset. - NewContext builds the proxy URL with ProxyHelper.asString() (scheme-less means http), looks up the exit IP through impit, and throws InvalidIP when the lookup fails instead of opening the context with the host's values. - No canvas seed is generated or sent (#721). noiseSeedsFromIdentity becomes audioSeedFromIdentity, and fpgen's constant navigator fields are no longer mapped. - The devicePixelRatio steps are ascending, so a tie goes to the lower step. - The two LeakWarning texts that named BrowserForge. - The README's note that Python's launch_server() ignored headless='virtual' is gone, because it no longer does. The golden fixtures are regenerated from #787's pythonlib. The generator now masks the fontconfig file name the way the test already did. The name hashes content that embeds the checkout path, so every regeneration from a different checkout used to rewrite 76 fixtures. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix: remove the glyph-spacing seed from the browser and the launcher anti-font-fingerprinting.patch added a seeded amount to every glyph advance, so that text widths differed per context. No real machine produces those widths: the same font on the same OS measures the same everywhere. So the noise was itself a fingerprint, measured in #779 at +1 px per ~100 glyphs plus fractional deltas on every measureText. #779 defaulted the seed to 0 and kept it as an opt-in, but an opt-in whose only effect is to become detectable is not worth carrying. Removed: - The browser side: - FontSpacingSeedManager and window.setFontSpacingSeed; - the HarfBuzz hook; - the plumbing that existed only to carry the context id down to the shaper: the userContextId on gfxTextRun, gfxShapedWord and the word-cache key, and the extra MakeTextRun argument in nsTextFrame, nsFontMetrics, MathML and canvas. The font group keeps its userContextId, which font-list-spoofing.patch uses to apply the per-context font list. Text is now shaped exactly as stock Firefox shapes it. - The fonts:spacing_seed key. The launcher had been sending 0 on every launch, plus a setFontSpacingSeed(0) call in every context's init script. - tests/patches/config-overrides.py, which tested only the spacing override. A pythonlib test now covers config_overrides with another key. timezone-spoofing, webrtc-ip-spoofing and window-setter-seal change only in context lines and the setter seal list. Every patch applies cleanly to a fresh tree, and the result builds. The settled decision is recorded as no-glyph-spacing-noise, with an automated check. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix: stock animations and speech by default; drop config keys that freeze live values Three behaviours a page could detect, changed in one breaking release: - **Animations run on stock timing.** no-css-animations.patch finished every finite animation at once by default, and any page could read it: `el.animate(frames, 1000).effect.getComputedTiming().duration` was 0, and a 500ms transition reported 0. Measured on v152.0.4-beta.31. The speedup is now an opt-in, `instantAnimations: True`, which raises a LeakWarning. disableInstantAnimations is gone. - **speak() on a spoofed voice works like a real voice.** It fired `error` after 3ms unless voices:fakeCompletion was set, and then start and end in the same tick. It now starts and ends after the text's duration at ~150 words per minute. Both voices:fakeCompletion keys are gone, and so is a debug line printed to stderr on every call. - **Keys removed:** - battery:* and window.scrollMinX/Y: Firefox keeps getBattery() and scrollMin* chrome-only, so no page could read them. - window.scrollMaxX/Y, screen.pageXOffset/pageYOffset, window.history.length and document.body.client*: each pinned a live value to a constant, so scrolling, navigating or re-laying out never changed it. fpgen.yml mapped pageYOffset, so about 15% of identities froze window.scrollY at a non-zero value. - The body keys' role as an undocumented alias for window.innerWidth/Height in browser-init and in the launcher. - MaskConfig::GetInt32Rect, which only the body keys used. New guards, both of which fail on v152.0.4-beta.31: tests/patches/animation-timing.py and tests/patches/spoofed-voice-speaks.py. The decisions are recorded as animations-run-on-stock-timing and spoofed-voices-speak. Every patch applies cleanly to a fresh tree, and the result builds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python)!: remove dead public API and make `list all --path` work Breaking changes: - Remove the exceptions UnknownProperty, InvalidDebugPort and MissingDebugPort. Nothing in the package raises them, so code catching them was catching nothing. - Remove the legacy `allow_webgl` keyword of launch_options(). Use `block_webgl=True`. The keyword now reaches Playwright as an unknown launch option and fails there instead of being silently consumed. `camoufox list all --path` accepted the flag and ignored it. It now prints the install path beside each installed build, as `camoufox list --path` already does for the installed tree. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(python)!: drop data the package never draws from voices.json shipped in the wheel, but no code in the package reads it: the voice draw uses voice-manifests.json and voice-uris.json. Its only readers are the TypeScript port's golden-fixture generator and data-sync script (typescript/scripts/golden/identity_golden.py, typescript/scripts/sync-identity-data.py), which live on another branch and will need a new source; the last copy is at 676fb3f:pythonlib/camoufox/voices.json. docs/per-context-patches.md described it as runtime data and now describes the files that are. webgl_data.db held two rows with zero weight on every OS ("Intel(R) HD Graphics 400, or similar" from "Intel Inc." and "Radeon R9 200 Series, or similar" from "ATI Technologies Inc."), left behind when their impossible macOS weights were zeroed. No draw can reach them. They are deleted with secure_delete so their blobs do not linger in free pages; the file is not vacuumed, so the other pages are unchanged. A new test requires every row to be drawable on at least one OS. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): warn whenever an identity falls back to a substitute value Several draws swallowed their failure and used something else, so an identity could ship with values the rest of it was not drawn to match and nobody would hear about it: - from_preset(): a failed font or voice draw used the preset's recorded list, or nothing, on any exception. - generate_context_fingerprint(): a failed font, voice or WebGL draw was `except Exception: pass`, leaving the browser's launch-time values. - _load_font_groups() / _load_font_bases(): an unreadable file became {}, i.e. no font additions or no OS-version base. - launch_options(): a failed font draw used every font in fonts.json, a failed voice draw used no voices, and a preset GPU missing from webgl_data.db was silently swapped for a drawn one (36 of the 397 bundled presets). Each site now catches only the errors its data can raise (OSError and ValueError for an unreadable or corrupt file, KeyError for a manifest with no entry for the OS, sqlite3.Error for the WebGL database) and emits a FallbackWarning. The text names what failed and what the identity uses instead, then gives a block to paste into an issue (camoufox, browser, OS and Python versions, the error, and the identity's user agent or GPU), asking the user to report it on GitHub. It shares LeakWarning's caller-frame attribution and its template lives in warnings.yml. The broad excepts had also been hiding a broken fixture: test_launch_environment's font and voice stubs did not accept `seed`, so every draw there raised and was swallowed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): give NewContext identities the browser's Firefox version NewContext() and AsyncNewContext() passed ff_version=None through to generate_context_fingerprint(), so a context's user agent kept the version fpgen drew (e.g. Firefox/146) while the browser underneath was 152. They now default ff_version to the major version of Playwright's Browser.version, which Juggler reports from MOZ_APP_VERSION_DISPLAY, so the UA always names the browser the page is actually talking to. An explicit ff_version still wins. The docstrings said each context gets "its own real fingerprint preset"; the default has been an fpgen draw, with a preset only when one is passed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): send an IPv6 WebRTC address to setWebRTCIPv6 The per-context init script passed every webrtc_ip, IPv6 included, to window.setWebRTCIPv4(), and never called setWebRTCIPv6(). An IPv6 address (given directly, or resolved as a proxy's exit IP) was stored as the context's IPv4 value and the IPv6 slot stayed empty. The script now picks the setter by address family, and an address that is neither raises InvalidIP instead of being passed through. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): stop pinning the page's scroll offset from fpgen fpgen.yml mapped the drawn window.pageYOffset (e.g. 528) to screen.pageYOffset, and the browser returns that value from scrollY on every read, so a page saw one scroll position forever whatever the user did. Real scroll offsets are live page state, not part of a device's fingerprint, so neither offset is mapped any more. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(python): stop checking a config key that no longer exists warn_manual_config() looked for navigator.languages, which was removed from settings/properties.json; validate_config() rejects it before the check could matter. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(python): close the WebGL database connection on every path sample_webgl raised its not-found and wrong-OS errors before reaching conn.close(), leaking a sqlite connection each time a preset named a GPU the database does not hold. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(data): drop the 23 presets whose GPU has no WebGL data A preset records only its GPU's name. The WebGL parameters, extensions and shader precision behind it have to come from somewhere, and for these 23 nothing Camoufox has describes the GPU: fpgen has never seen Firefox report it on that OS. So each launch paired the name with another device's parameters, a mismatch any WebGL fingerprinter can see. They were: - Windows on ARM (Adreno 650); - Direct3D 10-level GPUs (vs_4_0/vs_4_1); - "Generic Renderer"; - 945GM and GTX 480 on macOS; - nouveau/Mesa buckets on Linux; - one Linux preset pairing NVIDIA's proprietary vendor string with the nouveau renderer name. scripts/clean-fingerprint-data.py now applies the rule, via a shared fingerprints.firefox_gpus(), and test_shipped_data asserts it. 374 presets remain, and every OS keeps its presets. ROADMAP.md lists capturing WebGL data for these GPUs, which would bring them back. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): stop generating the glyph-spacing seed Mirrors676fb3f: the browser no longer has glyph-spacing noise, so the launcher sends no fonts:spacing_seed and the per-context init script no longer calls setFontSpacingSeed. config_overrides is now tested with audio:seed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): warn on instantAnimations; stop treating body keys as window size Mirrors the launcher half offb21b2e: instantAnimations raises the instant_animations LeakWarning (warnings.yml copied from pythonlib), the document.body.client* keys no longer count as window dimensions, and fpgen's pageYOffset is no longer mapped, so no identity freezes window.scrollY. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts)!: remove dead public API and make `list all --path` work Mirrors4616aa5: drop the UnknownProperty, InvalidDebugPort and MissingDebugPort exceptions (nothing raises them) and the legacy allow_webgl option (use block_webgl; allow_webgl now passes through to Playwright like any unknown option). `camoufox list all --path` prints each installed build's path, as `list --path` already did. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(ts)!: drop voices.json, which nothing draws from Mirrors9a42b6a. The voice draw reads voice-manifests.json and voice-uris.json; voices.json was only read by the golden generator and the data-sync script. The voice-URI golden now hashes the URI of every entry in voice-manifests.json, the list the draw actually picks from. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): warn whenever an identity falls back to a substitute value Mirrors165e68ffor the font and voice draws. Each fallback site in fromPreset(), generateContextFingerprint(), loadFontGroups(), loadFontBases() and launchOptions() now catches only the errors its data can raise and emits a FallbackWarning naming what failed, what the identity uses instead, and a block to paste into an issue (camoufox, browser, OS and Node versions, the error, the identity). The message is warnings.yml's `fallback` template, shared with pythonlib. Python's except clauses name builtin classes JavaScript lacks, so pycompat gains OSError, ValueError and KeyError twins and isPyError(): a Node system error counts as an OSError and JSON.parse's SyntaxError as a ValueError, as json.JSONDecodeError is. The voice draw now throws ValueError for a malformed entry and KeyError when the manifest has no macOS entry, as Python does. The WebGL fallback sites are left for the change that replaces the TS WebGL source. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): give NewContext identities the browser's Firefox version Mirrors46e5c8d: without an explicit ff_version, NewContext() kept the Firefox version fpgen drew, so a context's UA could name 146 on a 152 browser. It now defaults to the major version of Browser.version(). The option docs now say the default identity is an fpgen draw. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): send an IPv6 WebRTC address to setWebRTCIPv6 Mirrors0bbd152: the per-context init script passed every WebRTC IP to setWebRTCIPv4(), IPv6 included. It now picks the setter by address family and raises InvalidIP for an address that is neither. The init-script golden gains an IPv6 case. Also ports 7b43112's regression test: a drawn pageXOffset/pageYOffset is not carried into the config (the mapping went in fb21b2e's mirror). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(ts): stop checking a config key that no longer exists Mirrors480789a: warnManualConfig() looked for navigator.languages, which settings/properties.json no longer has; validateConfig() rejects it first. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(ts): sync the pruned presets and the properties.json fixture Copies fingerprint-presets*.json from pythonlib (40edebbdropped the 23 presets whose GPU has no WebGL data) and refreshes the launch fixture's copy of settings/properties.json, which lost the keys removed in676fb3fandfb21b2e. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(python): draw every identity's WebGL from fpgen WebGL vendor, renderer, context attributes, extensions, parameters and shader precisions, for WebGL1 and WebGL2, now come from fpgen's recorded Firefox devices instead of webgl_data.db, which is deleted with the camoufox/webgl/ package. camoufox/webgl.py: - webgl_for_gpu() traces `webgl` given Firefox, the OS and the GPU, then `webgl2` given the chosen `webgl` too, and draws each with one seeded random.Random. The GPU and the webgl value are pinned by their fpgen lookup index: a dict condition is flattened into leaves that overwrite each other, so only the renderer applied and Linux "Mesa" and "AMD" Radeon HD 3200 devices came back mixed. - sample_webgl_for_screen() draws the GPU of a generated identity from fpgen's per-OS weights, filtering out software rasterisers, GPUs the OS cannot report, discrete GPUs behind a netbook screen and the resistFingerprinting "Mozilla" mask before the weighted choice, so there is no rejection loop. An empty pool raises. - The draft/host-dependent extension filter moves over unchanged. A preset's GPU and a caller's webgl_config pair are looked up as given; a pair fpgen has never seen from Firefox on that OS raises instead of falling back to another GPU. generate_context_fingerprint no longer falls back to the host GPU when the draw fails. For 10 of the 15 (GPU, OS) pairs the two sources share, one of fpgen's records converts to exactly the database row on every value the browser reads. The other five rows (Linux R9 200 and Radeon HD 3200, macOS Intel HD, and two software rasterisers) are devices fpgen does not carry; those GPUs now present fpgen's recorded devices instead. The Linux GTX 980 row is kept as a test fixture. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs: say where WebGL comes from now that the database is gone The per-context guide, the fpgen.yml header and coherence's comments still named webgl_data.db and sample_webgl(). They now point at camoufox/webgl.py and fpgen. The guide also claimed presets carry WebGL parameters; they record only the vendor and renderer. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): regenerate the golden fixtures for the mirrored pythonlib changes Regenerates identity_golden.py and launch_golden.py output from this tree's pythonlib. launch_golden.py drops fonts:spacing_seed and document.body.clientWidth from its inputs, adds an instantAnimations scenario, and masks a FallbackWarning's report block in both launchers, since it names the host and the runtime. Two launch scenarios still differ: preset_windows_unknown_gpu and config_webgl_unknown_pair expect the FallbackWarning pythonlib now raises when a preset's GPU is missing from the WebGL data. That site belongs to the change replacing the TS WebGL source; the rest of each scenario matches. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(ts): draw every identity's WebGL from fpgen Mirrors0d859b3. src/webgl.ts is the twin of camoufox/webgl.py: webglForGpu() traces fpgen's webgl node given Firefox, the OS and the GPU, then webgl2 given the chosen webgl too; sampleWebglForScreen() first draws the GPU from fpgen's per-OS weights, filtered (no software rasteriser, no resistFingerprinting mask, a GPU the OS can report, no discrete GPU behind a netbook screen) before one weighted choice. Every draw is PyRandom.choices on one seeded instance, in Python's order. The GPU and webgl value are pinned by their fpgen lookup index, found from the value's stored JSON, which TraceResult now carries: re-serialising a parsed value would spell 2**64 differently from orjson. A preset's GPU and a caller's webgl_config are looked up as given and raise when fpgen has never seen them, instead of falling back to another GPU; generateContextFingerprint no longer swallows a failed draw. Removed with the old source: webgl/sample.ts, the numpy default_rng port (webgl/nprandom.ts), data-files/webgl_data.json and its export in sync-identity-data.py. The NumPy notice now covers the pairwise sum in locales.ts, the one NumPy port left. launchOptions now throws the pycompat ValueError where Python raises ValueError. Tests port test_webgl.py and the shipped-data check that every preset GPU has WebGL data. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): pin the fpgen WebGL draws against Python bit for bit identity_golden.py now records camoufox.webgl's output as sha256 of the exact orjson bytes (key order, int versus float and all): 1272 screen draws over every OS and seven screens (60 seeds each, plus four large seeds), webgl_for_gpu for every GPU fpgen records and every bundled preset GPU, the unknown-GPU and unknown-OS errors, and to_config's extension filter. The numpy golden keeps only np.sum, now checked against locales.ts. The renderer list the coherence golden walks comes from fpgen's traces. launch_golden.py takes its WebGL pairs from firefox_gpus(), and the unknown-GPU preset input is a GPU nobody records. The launch goldens are regenerated; the preset_windows_unknown_gpu and config_webgl_unknown_pair scenarios now expect Python's ValueError. The golden test no longer maps ValueError to Error. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(patches): a host's missing speech daemon no longer errors spoofed speech On a Linux host where speech-dispatcher cannot start, Firefox broadcasts synth-voices-error, and SpeechSynthesis answers it by firing `error` on every queued utterance. So a spoofed Windows voice errored about 11ms into speak() on any host without the daemon: the CI runners, and most servers. It passed only where the daemon runs. While Camoufox manages the voice list, the registry no longer forwards a host backend's error. The spoofed voices do not depend on the host's engine, and a Windows or macOS identity never raises one. The guard now makes the daemon unreachable itself, so it tests this case on every machine; on the previous build it fails every time. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci: stop skipping the two click tests that stock animation timing fixed test_wait_for_stable_position and test_timeout_waiting_for_stable_position were skipped with humanized travel time as the reason. The real cause was instant animations. Every finite animation finished at once, so the button Playwright waits on to stop moving never moved, and the click landed where upstream does not expect. With animations on stock timing both pass, and the skiplist audit flagged them as no longer failing. The entries go, and the counts in ci/README.md drop from 14 to 12. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(build-tester): accept 18 and 22 cores, as real hardware reports plausibleHWC's list of common core counts lacked 18 and 22 -- Intel Meteor Lake laptops (Core Ultra 5 125H, Core Ultra 7 155H), and 22 is in 8 recorded presets. build-tester draws random presets, so a run that picked one of the two Linux presets reporting 22 failed: about one run in eleven, on any pull request. A CI self-test now fails if the list rejects any core count pythonlib can present (the presets and PLAUSIBLE_CORE_COUNTS). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(guards): judge the query-cost probes on a median, not one sample stock-parity-probes timed each getter once. On a shared runner one GC pause or CPU-steal spike decided the verdict: navigator.hardwareConcurrency took 77 ms against a 50 ms allowance on the same restored build that passed the run before. Each pair is now timed five times, interleaved, and compared by median. The regressions these catch (a sync IPC per read, ~240 ms over the loop) cost extra on every read, so they move the median; verified by giving the getter a constant ~4 us of extra work per read -- 86 ms median, FAIL -- while the healthy build passes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(native): compare the whole fingerprint when two launches must differ test_two_browsers_get_different_fingerprints compared seven coarse values: UA, platform, screen size, core count, timezone and language. CI pins the timezone and language, and real machines share the rest: two draws of a common Mac (Firefox 152, MacIntel, 2560x1440, 8 cores) matched, and the test failed on a correct browser. It now reads the whole fingerprint a site computes, from a script in the page: - navigator values, screen and window geometry, device pixel ratio, timezone; - WebGL vendor, renderer, limits and extensions; - installed fonts, measured by width against the generic fallbacks; - voices, media-device counts, and an OfflineAudioContext hash. The page is served from an https URL Playwright fulfils locally, because mediaDevices exists only in a secure context. The page computes the result itself because the isolated world may not read audio sample data. The test then requires the fingerprints to differ, and the audio hash to differ on its own, since its noise is seeded per identity. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Update README to remove warning, camoufox is now actively maintained Camoufox will now be actively maintained and improved for the foreseeable future * ci(ts): hold the golden tests to live pythonlib, not a snapshot of it The typescript job ran the golden tests against the fixtures committed in typescript/tests/fixtures/. A pythonlib change that typescript/ did not mirror left those fixtures untouched, so the tests kept passing -- the opposite of what the job's comment promised. `ci.run_typescript --regenerate-golden` now rewrites the fixtures from the checkout's pythonlib before vitest runs, and CI passes it. The job runs Python 3.14 because pySum() reproduces sum() as 3.14 computes it; on 3.12 one crafted mixed int/float case differs in its last bit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci(ts): do not redraw fpgen's stats fixture on every run stats.json is thousands of random fpgen draws that the TS tests compare statistically. It changes with the pinned model, not with pythonlib, and redrawing it took eight of the typescript job's eleven minutes on a runner. The deterministic fpgen fixtures are still regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(native): measure crash growth from a warmed-up parent test_the_parent_stays_flat_across_content_crashes took its baseline right after launch. The parent's first context costs it 150-200 MB with no crash at all, so the warm-up counted as crash growth: 330-370 MB of the 400 MB allowance locally, and 469 MB on a CI runner, failing a PR that changes nothing in the browser. The baseline now follows one clean context; each crash still has to stay within the same allowance. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * refactor(ts): read pythonlib's data files instead of copying them typescript/src/data-files/ held byte-identical copies of eleven pythonlib data files (23k lines), kept in step by a sync script and a test that failed when a copy drifted. The TS launcher now reads them from pythonlib/camoufox/ when it runs from the repo, and `pnpm build` copies them into dist/data-files/ for the npm tarball, so what users install is unchanged. DATA_FILES in src/paths.ts is the one list; check-pack.mjs checks each is in the tarball. The essential-font lists were the one large table both ports hard-coded (~170 lines of Python, ~770 of TS). They move to pythonlib/camoufox/essential-fonts.json, which fingerprints.py and fingerprints.ts both read; gen-fonts-json.py --print-bases writes it and verify-fonts.py checks it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): record the golden fixtures from pythonlib on every run The goldens were pythonlib's output committed as 14k lines of fixtures, most of it the same input fingerprint pasted into ~80 launch scenarios, and CI already rewrote them from pythonlib before each run. They are now recorded by a vitest globalSetup (tests/golden-setup.ts) from the repo's .venv or $CAMOUFOX_PYTHON, in about 6 seconds, and git-ignored. A pythonlib change that typescript/ does not mirror fails `pnpm test` locally as well as in CI, and ci.run_typescript no longer needs --regenerate-golden. Committed inputs stay: launch/inputs.json, the bundle stubs, the addon, e2e/probe.js, and fpgen/stats.json (random draws tested statistically, which change with the pinned model, not with pythonlib, and take minutes to redraw). The one Python-version-sensitive case, sum() over mixed ints and floats, skips with a named prerequisite when the goldens come from Python < 3.14. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci(npm): give the publish job the pythonlib its tests record goldens from Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): NewContext hands Playwright the identity's user agent, DPR and timezone generateContextFingerprint already returns Playwright's JS option names; NewContext re-cased them, and camelCase() lowercases first, so userAgent, deviceScaleFactor and timezoneId became keys Playwright silently drops. navigator.userAgent was still spoofed at the C++ level, so the page probe matched Python's, but the HTTP User-Agent, the DPR and the timezone did not. The options now pass through as generated. NewContext also awaits ensureModel(): a browser from connect() or a custom executable never went through launchOptions(), which fetches the fpgen model, so an fpgen draw threw ModelNotInstalled where Python works. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): a failed browser download rejects instead of killing the process The download's file stream had no 'error' listener, so a failed write (disk full) was an uncaught 'error' event: Node exited before installVersioned()'s catch could remove the partial install and its temp directory, and the caller had nothing to catch. finished() now listens from the moment the stream is created, and webdl() surfaces an errored stream instead of writing into it. webdl() also waits for 'drain': it ignored write()'s return value, so on a slow disk the whole archive queued in memory. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): concurrent launches no longer read or inherit each other's CPU pin playwright-core spawns the browser from this process, so pin_cpu_cores narrows this process's own mask while a launch's browser starts. Python pins a separate driver and never sees its own mask change. Here, a second launch started in that window: - read the pinned mask as the host's cores (pinnedCoreCount, and the identity's hardwareConcurrency via availableParallelism()), and - if it did not pin, spawned its browser without the lock, inheriting the first launch's pin while reporting more cores. The host's core count is now read once, before this process first pins itself (cpu_affinity.hostCoreCount), and unpinned launches, launchServer included, take the pin lock once any launch in the process has pinned. With pin_cpu_cores off (the default) nothing waits. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ts): an Xvfb that fails to start throws CannotExecuteXvfb A spawn that fails after spawn() returns (EACCES, ENOENT) is an 'error' event on the child process, which had no listener: Node treated it as uncaught and exited instead of get() throwing. The child now always has a listener, readDisplayNumber() rejects with CannotExecuteXvfb on it, and the display pipe keeps an error listener after the read settles. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(scripts): gen-fonts-json refuses to write an incomplete essential-fonts.json An OS with no bases in the manifest was skipped, and the file written without its key; fingerprints.py and fingerprints.ts read every OS's list at import, so `import camoufox` then failed with a KeyError. The script now exits and keeps the existing file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * chore(release): pythonlib and the npm package to 0.5.7 @camoufox/camoufox 0.5.6 went to npm before the review fixes above, so they ship as 0.5.7; the two launchers are versioned in lockstep, and main already carries pythonlib changes from #787 that 0.5.6 on PyPI does not have. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): NewContext's HTTP User-Agent must match navigator.userAgent The e2e parity check read only navigator.userAgent, which the browser spoofs itself, so a context that dropped Playwright's userAgent option still matched Python. The probe server now records the request's User-Agent header. Against the NewContext before the fix, a context whose navigator said Windows sent the launch identity's Linux UA. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): draw the NewContext option-name test's preset from the v150 bundle getRandomPreset() without a Firefox version draws from the older bundle, where some Windows presets carry no devicePixelRatio, so the test failed on some draws in CI. Every v150 preset has one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(ts): wait for the headful e2e page to have focus, not one bringToFront() On a bare Xvfb, with no window manager, a single bringToFront() before goto() sometimes left the window unfocused, and Firefox holds enumerateDevices() until the document has focus, so the virtual-display probe timed out on some runs. Both launchers' probes now navigate first, then bring the page to the front until document.hasFocus() is true, and fail with that reason if it never is. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Camoufox
Camoufox is an open source anti-detect browser built for webscraping & AI agents. 🦊
Note
All of the latest documentation is available at camoufox.com.
Sponsors
View/Collapse All
Premium
|
|
NodeMaven: The most efficient proxy provider for Web Scrapping and Automation with the Highest Quality IP on the market. Why NodeMaven? • 99.9% uptime • ZIP Targeting • IP filtering: all proxies have fraud score <97% • No KYC required • Unique free tools: Proxy Bandwidth Checker, Meta Tag Checker, IP Lookup and others! Special codes for Camoufox users: • CAMOUFOX35 - 35% off to Mobile and Residential Proxies• CAMOUFOX40 - 40% off to ISP (Static) Proxies |
|
|
Need proxies for Camoufox? Use Node Proxy — a provider of datacenter, residential and mobile proxies offering high speed, security and near-100% uptime. Fully tested and supported in Camoufox. What sets them apart from other providers • 90+ IP score • HTTP + SOCKS5 — multiprotocol support • Discounts for retail customers • Full B2B support • Ethically sourced IP addresses • Special terms for Enterprise clients Want to support Camoufox? Just use my promo code CAMOUFOX — it gets you a 30% discount and supports the developer at the same time.Get started at node-proxy.com → |
|
|
Byteful is a UK-based web data infrastructure platform that provides ethically sourced residential, mobile, static residential (ISP) and datacenter proxies alongside API-first tools for web scraping, data collection, and AI-driven automation.
Processing tens of billions of requests per month for thousands of customers, Byteful powers browser-based AI agents, automation systems, and data workflows. It is a member of the Internet Watch Foundation and the Ethical Web Data Collection Initiative. Get 10% off Byteful Residential Bandwidth with the code: CAMOUFOX10 |
|
|
Layer3 Intel | See your proxies the way anti-bots see them Camoufox hides your browser. Your proxy IP is the part it can't hide. Layer3 Intel is the residential proxy detection engine used to catch proxy traffic — check whether the IPs your provider sells you are clean, or already known and flagged. • 🔍 Live threat score for any IP • 🗂️ Proxy pool membership - identify IP resellers • 📡 70M+ residential, mobile & ISP proxy IPs tracked across 200+ providers • ⚡ <40ms API responses — vet IPs inline before your scraper uses them Check your IPs: https://layer3intel.com |
Tools & Services
|
Scrapfly is an enterprise-grade solution providing Web Scraping API that aims to simplify the scraping process by managing everything: real browser rendering, rotating proxies, and fingerprints (TLS, HTTP, browser) to bypass all major anti-bots. Scrapfly also unlocks the observability by providing an analytical dashboard and measuring the success rate/block rate in detail. |
|
Clover Labs is a Toronto based venture studio building AI agents for growth and distribution. |
|
|
SerpApi, a web search API to scrape Google and other search engines with a simple API. |
|
|
Web data that survives the anti-bots. Crawlbase gives developers and AI teams reliable data at scale: a 99% success-rate Crawler, Crawling API, Smart AI Proxies, and Web MCP Server that get through, so your scrapers and agents don't break. You build, we handle the infrastructure. Get 15% off your first 3 months with code CAMOUFOX → crawlbase.com |
|
|
Scrappey is a Web Scraping API that only charges successful scrapes with pay as you go - no subscriptions. Scrape complex sites. Residential proxies included, no hidden proxy fees, or expiring balances. One API for direct HTTP, full-browser rendering, JavaScript-heavy pages, screenshots, sessions, 30+ browser actions and 200+ concurrent sessions at a time - trusted by 1000+ developers and AI agents. Get 10% off with code CAMOUFOX. |
|
|
Cloro is a SERP and AI search API. Get structured results from Google, ChatGPT, Perplexity, Gemini, Copilot and Grok. |
Proxy Providers
Camoufox is intended to be used with rotating proxies (preferably residential IPs). Check out these providers:
|
|
🚀 Camoufox × ProxyEmpire Running Camoufox? Your proxy layer decides whether you scale — or get blocked. ProxyEmpire delivers: • 🌍 30M+ Residential IPs (170+ countries) • 📱 4G/5G Mobile Proxies • 🔄 Rotating & Sticky Sessions • ⚡ Unlimited Concurrent Sessions • 🎯 Precise geo-targeting • HTTP, HTTPS & SOCKS5 Support Built for scraping, automation, and high-stealth workflows. 🔥 Exclusive Offer - Use code Camoufox30 Get 30% recurring discount (not just first month). Upgrade your proxies. Reduce bans. Scale properly |
|
|
RapidProxy - Power Your Data with Premium Proxies. 🎁 Try proxies for free + Use code RAPID10 for 10% OFF Why Choose RapidProxy? • 🌍 90M+ IPs in 200+ countries & regions • ♾️ No expiration on traffic — use anytime, no pressure • 🔥 Unlimited concurrency for maximum performance • 💰 Starting from just $0.65/GB — built for scale • 📍 City-level targeting for precise geo access • 🔄 Flexible session control tailored to your needs Don’t miss out — start your free trial today and experience fast, stable, and scalable proxy performance with RapidProxy. |
|
|
Swiftproxy - High-Performance Residential Proxies for Scalable Data Collection Built for developers who need reliable, anti-detection proxy infrastructure. Swiftproxy delivers stable connections, high success rates, and flexible control for large-scale scraping and automation. • 🌍 195+ locations with ethically sourced residential IPs • 🔄 Rotating & sticky sessions with precise geo-targeting • ⚡ Optimized for anti-ban & high success rate • 🔌 HTTP / HTTPS / SOCKS5 support • 🧪 Free 500MB trial for testing • 💸 Special discount code for Camoufox users: PROXY90 - 10% Best for: Web scraping, automation, multi-accounting, and large-scale data extraction |
|
|
MangoProxy is a Residential, ISP, Mobile and Datacenter proxy service designed for professional tasks where stability, speed, and anonymity matter. Use code DAIJRO for 8% OFF ISP Static Proxies |
|
|
Proxidize | Mobile and Residential Proxies for Camoufox Running Camoufox at scale? Your browser setup is only half the stack. Your proxy layer matters too. Proxidize provides mobile and residential proxies built for scraping, browser automation, SEO monitoring, AI agents, and data collection workflows. Why Proxidize? • Real 4G and 5G mobile proxies • Residential proxies in 195+ countries • Rotating and sticky sessions • City-level and carrier targeting • Unlimited concurrency • HTTP(S), SOCKS5, and UDP over SOCKS support • No hardware or DIY setup required Built for teams that need reliable proxy infrastructure without managing devices, servers, or proxy rotation themselves. Special offer for Camoufox users: Use code CAMOUFOX20 for 20% off. Start now: https://proxidize.com |
|
|
NiuProxy | Rotating Residential Proxies from $0.35/GB NiuProxy provides residential, ISP, mobile, and datacenter proxies for scraping, browser automation, SEO, AI agents, and data collection. Why NiuProxy? • Residential proxies from $0.35/GB • ISP proxies from $3/IP • Mobile proxies from $1.5/GB • Datacenter proxies from $0.5/GB • HTTP(S) & SOCKS5 support • Flexible geo targeting and sessions • Alipay, USDT, cards, Google Pay & Apple Pay Special offer for Camoufox users: Use code PAY2 for 10% off your recharge. Start now: https://niuproxy.com |
|
|
🔍 Camoufox × Thordata Real Residential IPs for Smarter AI Agents & Automation With 100M+ residential IPs, Thordata helps your scraper access the web through real user IPs across 195+ countries. Target specific locations with precision — including city, ISP, and ASN-level targeting — so your Camoufox automation runs with a more authentic network identity. • 🔄 Rotating & Sticky Sessions (up to 90 minutes) • ⚡ 99.99% uptime with unlimited concurrent sessions • 🌍 Global residential coverage for AI agents, scraping, and automation workflows 🎁 Exclusive for Camoufox users: Get free trial traffic after signup + use code Camoufox for 10% OFF. Start your free trial with Thordata |
|
|
Webshare gives you instant access to a proxy pool of 80M+ ethically-sourced IPs across 195+ countries, with rotating residential, static ISP, and datacenter options plus a full API. It includes a 100+ Gbps backbone, country/city/state/ZIP/ASN-level targeting, and requires no credit card to start. 🏷️ Get 20% OFF your first purchase with promo code CAMOUFOX20
|
|
Roam — Residential & static residential proxies, pay-as-you-go per GB. Use code CAMOUFOX15 for 15% extra credit on your first top-up. |
Introduction
Camoufox is a Firefox fork engineered for web scraping and AI agents. It is headless, undetectable, and optimized to run at scale. Every run gets a fresh identity drawn from the real-world distribution of devices, so it blends into normal traffic instead of standing out.
Highlights
- Built for AI agents 🤖
- Minimal, debloated Firefox - fast to launch, cheap to run
- Drop-in Playwright compatibility from Python and JavaScript/TypeScript
- Invisible to anti-bot systems so you can run your agent cluster locally or in the cloud without being flagged
- Undetectable by design 🎭
- Page automation hidden from JavaScript inspection. See the stealth page for more details.
- Fingerprint injection & rotation (without JS injection!)
- All navigator properties (device, OS, hardware, browser, etc.) ✅
- Screen size, resolution, window, & viewport properties ✅
- Geolocation, timezone, locale, & Intl spoofing ✅
- WebRTC IP spoofing at the protocol level ✅
- Voices, speech playback rate, etc. ✅
- And much, much more!
- Anti Graphical fingerprinting
- WebGL parameters, supported extensions, context attributes, & shader precision formats ✅
- Font spoofing & anti-fingerprinting ✅
- Optimized for automation
- Human-like mouse movement 🖱️
- Blocks & circumvents ads 🛡️
- Optional instant animations (
instantAnimations), so Playwright never waits on one 💨
- Debloated & optimized for memory efficiency ⚡
- PyPI and npm packages for updates & auto fingerprint injection 📦
- Stays up to date with the latest Firefox version 🕓
Fingerprint Injection
In Camoufox, data is intercepted at the C++ implementation level, making the changes undetectable through JavaScript inspection.
To spoof individual fingerprint properties, pass a JSON containing properties to spoof to the Python or TypeScript interface:
>>> with Camoufox(config={"property": "value"}) as browser:
Config data not set by the user is populated from fpgen, a model of the statistical distribution of device characteristics in real-world traffic. The assembled identity is then checked for coherence, so parts that are each plausible cannot combine into a machine that does not exist.
Usage
Camoufox is compatible with your existing Playwright code. You only have to change your browser initialization.
Python, sync API
from camoufox.sync_api import Camoufox
with Camoufox() as browser:
page = browser.new_page()
page.goto("https://example.com")
Python, async API
from camoufox.async_api import AsyncCamoufox
async with AsyncCamoufox() as browser:
page = await browser.new_page()
await page.goto("https://example.com")
JavaScript / TypeScript
import { Camoufox } from "@camoufox/camoufox";
const browser = await Camoufox({ headless: true });
const page = await browser.newPage();
await page.goto("https://example.com");
await browser.close();
[Python installation & usage] · [TypeScript package]
Capabilities
Below is a list of patches and features implemented in Camoufox.
Fingerprint spoofing
- Navigator properties spoofing (device, browser, locale, etc.)
- Support for emulating screen size, resolution, etc.
- Spoof WebGL parameters, supported extensions, context attributes, and shader precision formats.
- Spoof inner and outer window viewport sizes
- Spoof AudioContext sample rate, output latency, and max channel count
- Spoof device voices & playback rates
- Spoof the amount of microphones, webcams, and speakers available.
- Network headers (Accept-Languages and User-Agent) are spoofed to match the navigator properties
- WebRTC IP spoofing at the protocol level
- Geolocation, timezone, and locale spoofing
- etc.
Stealth patches
- Avoids main world execution leaks. All page agent javascript is sandboxed
- Avoids frame execution context leaks
- Fixes
navigator.webdriverdetection - Fixes Firefox headless detection via pointer type (#26)
- Removed potentially leaking anti-zoom/meta viewport handling patches
- Uses non-default screen & window sizes
- Re-enable fission content isolations
- Re-enable PDF.js
- Other leaking config properties changed
- Human-like cursor movement
Anti font fingerprinting
- Automatically uses the correct system fonts for your User Agent
- Bundled with Windows, Mac, and Linux system fonts
- No glyph-spacing noise: measured text widths are the ones the same font gives on a real machine
Playwright support
- Custom implementation of Playwright for the latest Firefox
- Various config patches to evade bot detection
Debloat/Optimizations
- Stripped out/disabled many, many Mozilla services. Runs faster than the original Mozilla Firefox, and uses less memory (200mb)
- Patches from LibreWolf & Ghostery to help remove telemetry & bloat
- Debloat config from PeskyFox, LibreWolf, and others
- Speed & network optimizations from FastFox
- Animations run on stock timing;
instantAnimations: Truefinishes them at once, at the cost of being detectable - Minimalistic theming
- etc.
Addons
- Load Firefox addons without a debug server by passing a list of paths to the
addonsproperty - Added uBlock Origin with custom privacy filters
- Addons are not allowed to open tabs
- Addons are automatically enabled in Private Browsing mode
- Addons are automatically pinned to the toolbar
- Fixes DNS leaks with uBO prefetching
Python & TypeScript Interfaces
- Automatically generates & injects unique device characteristics into Camoufox based on their real-world distribution
- WebGL fingerprint injection & rotation
- Uses the correct system fonts and subpixel antialiasing & hinting based on your target OS
- Avoid proxy detection by calculating your target geolocation, timezone, & locale from your proxy's target region
- Calculate and spoof the browser's language based on the distribution of language speakers in the proxy's target region
- Remote server hosting to use Camoufox with other languages that support Playwright
- Built-in virtual display buffer to run Camoufox headfully on a headless server
- Toggle image loading, WebRTC, and WebGL
- etc.
Note
Camoufox does not fully support injecting Chromium fingerprints. Some WAFs (such as Interstitial) test for Spidermonkey engine behavior, which is impossible to spoof.
Stealth Overview
How Camoufox hides its automation library
In Camoufox, all of Playwright's internal Page Agent's code is sandboxed and isolated. This makes it impossible for a page to detect the presence of Playwright through Javascript inspection.
Normally, Playwright injects some JavaScript into the page such as window.__playwright__binding__ and to perform actions like querying elements, evaluating javascript, or running init scripts, which can be detected by websites. In Camoufox, these actions are handled in an isolated scope outside of the page. In other words, websites can no longer "see" any JavaScript that Playwright would typically inject. This prevents traces of Playwright altogether.
However, even with hiding its automation library, Camoufox is not immune to inconsistencies in fingerprint rotation. This still requires maintenance to spot and fix.
Page Interactions
Anti-bot systems also run client-side scripts to monitor your behavior. For example, they look for patterns in mouse movements, clicks, scrolling, and the timing between actions.
Every dot above is a mousemove event the page received from six page.mouse.move() calls, replayed at the speed it arrived. Close dots mean the hand slowed down. scripts/cursor-demo.py regenerates the figure from a build.
Camoufox does not draw its cursor paths. With humanize=True it uses Cursory by Vinyzu, which holds 2357 mouse movements recorded from real people: it picks a recording whose direction, distance and wander suit the move being made, morphs it onto the requested start and end points, and replays it with that recording's own timing — pauses, overshoots and all.
That last part matters as much as the shape. Camoufox previously walked a Bézier curve through two random knots and emitted a point every 10ms. Both halves of that are tells: an analytic curve sampled at a fixed rate has velocity and jerk profiles that separate cleanly from a hand's, and the acceleration came entirely from one easing function, so every movement Camoufox ever made sped up and slowed down the same way. A replayed recording has neither property.
Camoufox ships cursory-js, a TypeScript port of Cursory, vendored into Juggler at additions/juggler/input/cursory/. It reproduces the Python original bit for bit, so a path can be reproduced against pip install cursory. Cursory is LGPLv3-or-later, not MPL-2.0 like the rest of the browser; its licence and full provenance are in additions/juggler/input/cursory/NOTICE.
However, this isn't perfect. It may still be detected with sophisticated enough analysis. (WIP for the future)
How Camoufox rotates identities
AI agents need to operate across many sessions without getting flagged or rate-limited. Rotating your IP address isn't enough — every browser session carries thousands of signals that create a unique fingerprint. A website can see your OS, GPU, screen resolution, fonts, timezone, and more. If those signals are inconsistent or unusual, you get blocked.
Market Share Distribution
Even if you are rotating your IP for each running bot instance, web access firewalls can still use machine learning to analyze incoming web traffic to detect if it's abnormal. If the Linux market share was 5%, then suddenly it's 20%, it's a red flag. They will unconditionally require all Linux users to complete a captcha.
Camoufox draws identities from fpgen, a Bayesian network trained on live traffic, so each device characteristic appears about as often as it does in the real world, and in the combinations real devices produce.
How can Camoufox be detected?
Camoufox can spoof fingerprints with a correct market share. However, fingerprints must also be internally consistent. A Windows user agent with an Apple M1 GPU, a MacOS user agent with a Windows DirectX renderer, and a mobile device with a desktop screen resolution are all impossible, and will be flagged for being suspicious.
Every drawn identity passes a coherence check (pythonlib/camoufox/coherence.py) that rejects impossible combinations before launch. But of the thousands of datapoints that must agree with each other, Camoufox doesn't always get every one right. Anti-bot providers test Camoufox over and over again to find even 1 unique inconsistency, then they immediately update their background scripts to test for it.
How does Camoufox compare to other solutions?
JavaScript-based solutions
In the past, developers tried injecting JavaScript to spoof these values, but it doesn't work reliably since JavaScript can't spoof everything. Incomplete coverage causes inconsistent fingerprints. For example, an anti-bot system will flag you if your network request's User Agent doesn't match your navigator's User Agent.
Additionally, all injected JavaScript is detectable in some way. Anti-bot systems can check if Object.getOwnPropertyDescriptor reveals an overwritten property, if a function's toString() no longer returns [native code] (revealing it was hijacked), or if data in the window context doesn't match the worker thread context. Workarounds only take you so far, but there will always be a way to detect JS injection if you search deep enough.
Camoufox's approach
Since Camoufox intercepts calls in the browser's C++ implementation level, all of the hijacked objects and properties appear native. There is no JavaScript hijacking to be detected.
Camoufox also generates consistent and believable fingerprints with fpgen and its coherence check. However, this can still be detected by complex fingerprint detection methods like mismatching data (as described earlier).
CDP-based libraries
CDP (Chrome DevTools Protocol) is an automation protocol built into Chromium and Firefox. However, CDP makes no effort to hide the fact that it's an automation protocol and exposes much of its functionality in the page scope. Some common methods are checking if navigator.webdriver is true, catching it reading the stack debugger, checking for variables that ChromeDriver injects into the document object for internal communication, and more.
Camoufox's approach
While Playwright uses CDP to control Chromium, it uses Juggler for Firefox. Juggler is a custom protocol developed before Firefox supported CDP (original repo). It is a distinct module within Firefox, and not part of its core browser. This makes it easier to edit and control what's revealed to the page.
Camoufox patches Juggler to give it its own isolated "copy" of the page to work with. Playwright can read and edit its own version of the page freely. Everything appears to work normally to it, but the real page is completely unaffected by these changes. The page also can't detect when things are being read (through tricks like hijacking getters) or listeners being added to watch elements.
Additionally, Juggler sends its inputs directly through the Firefox's original user input handlers, meaning they are handled the exact same way as if you were using the browser normally. Camoufox also patches Firefox's headless mode to appear the same as if it were running in a normal window. But as a fallback, the Python library can run Camoufox in a virtual display if headless mode ever leaks.
Build System
Warning
The content below is intended for those interested in building & debugging Camoufox. For usage instructions, see pythonlib or typescript.
Overview
Here is a diagram of the build system, and its associated make commands:
graph TD
FFSRC[Firefox Source] -->|make fetch| REPO
subgraph REPO[Camoufox Repository]
PATCHES[Fingerprint masking patches]
ADDONS[uBlock Origin]
DEBLOAT[Debloat/optimizations]
SYSTEM_FONTS[Win, Mac, Linux fonts]
JUGGLER[Patched Juggler]
end
subgraph Local
REPO -->|make dir| PATCH[Patched Source]
PATCH -->|make build| BUILD[Built]
BUILD -->|make package-linux| LINUX[Linux Portable]
BUILD -->|make package-windows| WIN[Windows Portable]
BUILD -->|make package-macos| MAC[macOS Portable]
end
This was originally based on the LibreWolf build system.
Build CLI
Warning
Camoufox's build system is designed to be used in Linux. WSL will not work!
First, clone this repository with Git:
git clone --depth 1 https://github.com/daijro/camoufox
cd camoufox
Next, build the Camoufox source code with the following command:
make dir
Before bootstrapping, install the system build dependencies with the helper
script. It detects your platform and installs everything the build needs
(Python ≥ 3.11, Rust, aria2, p7zip, msitools, wget, sqlite, and
the core build tools) using the appropriate package manager — Homebrew on macOS,
or apt/dnf/pacman on Linux:
bash scripts/install-deps.sh
Note
The dependency installer has so far only been tested on macOS.
After that, you have to bootstrap your system to be able to build Camoufox. You only have to do this one time. It is done by running the following command:
make bootstrap
Finally you can build and package Camoufox the following command:
python3 multibuild.py --target linux windows macos --arch x86_64 arm64 i686
For new builds, i686 is supported only for Windows. Unsupported target/architecture combinations are skipped.
CLI Parameters
Options:
-h, --help show this help message and exit
--target {linux,windows,macos} [{linux,windows,macos} ...]
Target platforms to build
--arch {x86_64,arm64,i686} [{x86_64,arm64,i686} ...]
Target architectures to build for each platform
--bootstrap Bootstrap the build system
--clean Clean the build directory before starting
Example:
$ python3 multibuild.py --target linux windows macos --arch x86_64 arm64
Using Docker
Camoufox can be built through Docker on all platforms.
- Create the Docker image containing Firefox's source code:
docker build -t camoufox-builder .
- Build Camoufox patches to a target platform and architecture:
docker run -v "$(pwd)/dist:/app/dist" camoufox-builder --target <os> --arch <arch>
How can I use my local ~/.mozbuild directory?
If you want to use the host's .mozbuild directory, you can use the following command instead to run the docker:
docker run \
-v "$HOME/.mozbuild":/root/.mozbuild:rw,z \
-v "$(pwd)/dist:/app/dist" \
camoufox-builder \
--target <os> \
--arch <arch>
Docker CLI Parameters
Options:
-h, --help show this help message and exit
--target {linux,windows,macos} [{linux,windows,macos} ...]
Target platforms to build
--arch {x86_64,arm64,i686} [{x86_64,arm64,i686} ...]
Target architectures to build for each platform
--bootstrap Bootstrap the build system
--clean Clean the build directory before starting
Example:
$ docker run -v "$(pwd)/dist:/app/dist" camoufox-builder --target windows macos linux --arch x86_64 arm64 i686
Build artifacts will now appear written under the dist/ folder.
Working on patches
make dir leaves camoufox-<version>-<release>/ as a git repository with every patch applied. A patch is a diff against a checkpoint in that repository:
# A new patch
make dir # apply every existing patch
make first-checkpoint # mark the starting point
# ...edit files in camoufox-*/, test with `make build` and `make run`...
make diff > patches/my-change.patch
# An existing patch
make dir
make workspace ./patches/x.patch # unapply x, checkpoint, reapply x
# ...edit...
make diff > patches/x.patch
make diff shows only tracked files, so git add -N <file> any new file first. make workspace needs every later patch to leave the hunks of x.patch alone. make patch and make unpatch apply or reverse one patch, and make revert resets the tree to unpatched Firefox.
Then run the suites that cover what you changed. CONTRIBUTING.md says which ones, and ci/README.md has the whole pipeline.
Leak Debugging
This is a flow chart demonstrating my process for determining leaks without deobfuscating WAF Javascript. The method incrementally reintroduces Camoufox's features into Firefox's source code until the testing site flags.
This process requires a Linux system and assumes you have Firefox build tools installed (see here).
See flow chart...
flowchart TD
A[Start] --> B[Does website flag in the official Firefox?]
B -->|Yes| C[Likely bad IP/rate-limiting. If the website fails on both headless and headful mode on the official Firefox distribution, the issue is not with the browser.]
B -->|No| D["Run make ff-dbg(1) and build(2) a clean distribution of Firefox. Does the website flag in Firefox **headless** mode(4)?"]
D -->|Yes| E["Does the website flag in headful mode(3) AND headless mode(4)?"]
D -->|No| F["Apply config.patch(5), then rebuild(2). Does the website still flag(3)?"]
E -->|No| G["Enable privacy.resistFingerprinting in the config(6). Does the website still flag(3)?"]
E -->|Yes| C
G -->|No| H["In the config(6), enable FPP and start omitting overrides until you find the one that fixed the leak."]
G -->|Yes| I[If you get to this point, you may need to deobfuscate the Javascript behind the website to identify what it's testing.]
F -->|Yes| K["Apply playwright/0-playwright.patch(5), then rebuild. Does it still flag?"]
F -->|No| J["Omit options from camoufox.cfg(6) and rerun(3) until you find the one causing the leak."]
K -->|No| M[Juggler needs to be debugged to locate the leak.]
K -->|Yes| L[The issue has nothing to do with Playwright. Apply the rest of the Camoufox patches one by one until the one causing the leak is found.]
M --> I
Cited Commands
| # | Command | Description |
|---|---|---|
| (1) | make ff-dbg |
Setup vanilla Firefox with minimal patches. |
| (2) | make build |
Build the source code. |
| (3) | make run |
Runs the built browser. |
| (4) | make run args="--headless https://test.com" |
Run a URL in headless mode. All redirects will be printed to the console to determine if the test passed. |
| (5) | make patch ./patches/<name>.patch |
Apply one patch. make unpatch reverses it. |
| (6) | make edit-cfg |
Edit camoufox.cfg in the default system editor. |
Licensing
- The browser (
patches/,additions/,settings/, and the build system) is MPL-2.0, the licence of the Firefox source it modifies. The vendored Cursory trajectories are LGPLv3-or-later (additions/juggler/input/cursory/NOTICE). - The launchers are MIT: the Python package (
pythonlib/LICENSE) and the TypeScript package (typescript/LICENSE). The TypeScript package also contains ports of fpgen, CPython'srandomand NumPy's random generators; their notices are intypescript/THIRD_PARTY_NOTICES.md.
Thanks
Debloating & references:
- LibreWolf: Debloat patches & build system inspiration
- BetterFox: Speed and debloat preferences
- Ghostery: Debloat reference (disable onboarding)
Web scraping & testing:
- Vinyzu/cursory: The recorded human mouse trajectories behind
humanize=True, vendored via cursory-js (LGPLv3-or-later — seeadditions/juggler/input/cursory/NOTICE) - riflosnake/HumanCursor: The Bézier cursor algorithm Camoufox used before Cursory
- scrapfly/fingerprint-generator (fpgen): The device distribution identities are drawn from
- CreepJS, Browserleaks, BrowserScan - Valuable leak testing sites
UI theming:
- Jamir-boop/minimalisticfox: Inspired Camoufox's minimal css theming (link)