katago-cuda

maintainer chengscott · 2 votes · scanned 2026-08-03 00:08:14.047287
LOW
View on AUR ↗
Why flagged The package downloads a model file from a non-whitelisted host, but it is a data file (neural network weights) that is not executed; the actual binary is built from source, and the model is used only by the locally compiled katago engine.

Triggered rules

LOW AI review downgraded a static finding llm_review

The static rules flagged this MEDIUM, but an AI model (qwen/qwen3-235b-a22b-07-25) reviewed the full PKGBUILD and judged it LOW (confidence 95%): The package downloads a model file from a non-whitelisted host, but it is a data file (neural network weights) that is not executed; the actual binary is built from source, and the model is used only by the locally compiled katago engine.

1 higher static finding superseded - not the current verdict (shown for transparency)
MEDIUM source=() URL on a non-standard host source_untrusted_domain

One or more source=() URLs point to a host outside the trusted allowlist (github.com, gitlab.com, codeberg.org, pypi.org, …).

  • PKGBUILD:19 "https://media.katagotraining.org/uploaded/networks/models/kata1/kata1-b28c512nbt-adam-s11165M-d5387M.bin.gz"

PKGBUILD

1 offending line(s) highlighted
1# Maintainer: Scott Cheng <aur@chengscott.io>
2# Contributor: Alexander F. Rødseth <xyproto@archlinux.org>
3# Contributor: flovo <flovo-git-commit@f-voegele.de>
4
5pkgname=katago-cuda
6_pkgname=katago
7pkgver=1.16.4
8pkgrel=1
9pkgdesc='Baduk engine with no human-provided knowledge'
10arch=(x86_64)
11provides=($_pkgname)
12conflicts=($_pkgname)
13url='https://github.com/lightvector/KataGo'
14license=(MIT)
15depends=(libzip)
16makedepends=(cmake cuda cudnn ninja)
17source=("${pkgname}-${pkgver}.tar.gz"::"${url}/archive/refs/tags/v${pkgver}.tar.gz"
18 # Strongest confidently-rated network
19 "https://media.katagotraining.org/uploaded/networks/models/kata1/kata1-b28c512nbt-adam-s11165M-d5387M.bin.gz"
20)
21b2sums=('7356e1c71c4f9a6c3cfbb9e4bccc07b21ce1d84265fd4ade11be65e0eb3dbadb50e9e52bd64f1266023cc084bf52aadbd15d76495cab29eed61d398473d8e82f'
22 '4aa7a435dc5a9bcd1abc3d30baa17530382138d4975cfcedd43aa44e7273c1ae558627ba70e0c9ffe34159d00372c70566fc56a05538781d447fed4d0b7d2eef'
23)
24_model_source=("${source[@]:1}")
25_model=("${_model_source[@]##*/}")
26noextract=("${_model[@]}")
27
28prepare() {
29 cd "KataGo-${pkgver}/cpp"
30 sed -i '/if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12\.8)/c\
31if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 13.0)\
32 set(CMAKE_CUDA_ARCHITECTURES 75 80 86 87 90 120)\
33elseif(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8)' CMakeLists.txt
34}
35
36build() {
37 cmake \
38 -S "KataGo-${pkgver}/cpp" \
39 -B build \
40 -D CMAKE_BUILD_TYPE=Release \
41 -D CMAKE_INSTALL_PREFIX=/usr \
42 -D USE_BACKEND=CUDA \
43 -D NO_GIT_REVISION=1 \
44 -G Ninja
45 cmake --build build
46}
47
48package() {
49 install -Dm755 build/katago "${pkgdir}/usr/bin/katago"
50 install -Dm644 "KataGo-${pkgver}/LICENSE" "${pkgdir}/usr/share/licenses/${_pkgname}/LICENSE"
51 # install models
52 for mi in "${_model[@]}"; do
53 install -Dm644 "${mi}" "${pkgdir}/usr/share/${_pkgname}/${mi}"
54 done
55}
56

Scan history

Scanned at (UTC)SeverityRules
2026-08-03 00:08:14 LOW 2
2026-08-02 00:16:08 LOW 2
2026-08-01 00:11:18 LOW 2
2026-07-31 00:14:10 LOW 2
2026-07-30 00:17:23 LOW 2
2026-07-29 00:25:53 LOW 2
2026-07-28 00:07:28 LOW 2
2026-07-27 00:24:32 LOW 2
2026-07-26 00:07:32 LOW 2
2026-07-25 00:13:44 LOW 2
2026-07-24 00:02:28 LOW 2
2026-07-23 00:14:47 LOW 2
2026-07-22 00:29:32 LOW 2
2026-07-21 00:24:15 LOW 2
2026-07-20 00:19:49 LOW 2
2026-07-19 00:17:08 LOW 2
2026-07-18 00:14:48 LOW 2
2026-07-17 00:06:16 LOW 2
2026-07-16 00:05:41 LOW 2
2026-07-15 00:09:25 LOW 2

Report a package

Reports go to the AURWatch maintainer (one person) and are read by hand. No login required.

0 / 4000
Your suggestion