SkCodecFuzzer is a small utility for testing the security and reliability of C/C++ image codecs supported by the Skia graphics library. In Android, these parsers are reachable through standard interfaces such as BitmapFactory and BitmapRegionDecoder and execute in the context of local apps (not a sandboxed media server), which exposes them to remote attacks via MMS, chat apps, emails etc. While the decoders available in Android by default (bmp, png, jpeg, gif, ...) are all open-source and already subject to extensive fuzzing, there may exist additional lesser-known, proprietary codecs added by device manufacturers. Such codecs aren't put under the same scrutiny due to their closed-source nature, and they may go unaudited, non-fuzzed or even completely unnoticed for many years. A notable example is the Qmage format (.qmg file extension), which was introduced in Skia on Samsung Android phones in late 2014, but was only recognized as an attack surface at the end of 2019. It has been used as the container for image resources in built-in Samsung APKs and themes in some (but not all) firmwares. The loader in this repository was used by Google Project Zero to run Qmage fuzzing at scale in January 2020, resulting in the uncovering of 5218 unique crashes, including hundreds of memory corruption issues (buffer overflows, use-after-free's etc.). They were subsequently reported to Samsung on January 28 as issue #2002 in the PZ bug tracker, and fixed by the vendor in May 2020. For additional context and more information about .qmg files, we recommend to refer to that tracker entry as it aims to explain our effort in great detail. The purpose of this harness is to link to Android's precompiled ARM(64) Skia libraries (libhwui.so or libskia.so on older versions) and use its SkCodec class to load an input file, in the same way that BitmapFactory::doDecode decodes images on real Android devices. It can run on both physical phones with ARM CPUs and in an emulated qemu-aarch64 environment on any host CPU of choice, enabling effective parallelized fuzzing.