blob: 188e20465d20b9046e11f4feb658774d68e01dcf (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
|
# Maintainer: SkillOpt Team <https://github.com/microsoft/SkillOpt/issues>
pkgbase=python-skillopt
pkgname=('python-skillopt' 'python-skillopt-webui')
pkgver=0.2.0
pkgrel=1
pkgdesc="Agentic Skill Optimization via Reflective Training Loops"
arch=('any')
url="https://github.com/microsoft/SkillOpt"
license=('MIT')
install=python-skillopt.install
makedepends=(
'python-build'
'python-installer'
'python-setuptools'
'python-wheel'
# Runtime deps are also listed in makedepends so namcap's split-package
# check passes: each sub-package can be built independently and must find
# its full transitive closure on the build host.
'python-azure-core'
'python-azure-identity'
'python-httpx'
'python-numpy'
'python-openai'
'python-openpyxl'
'python-yaml'
'python-gradio'
# Note: omegaconf, pytorch, ray are NOT in makedepends. They are lazy-imported
# only when the user activates the alfworld extra, so they live in optdepends
# only. Listing them here would force every `makepkg -s` user to install ~2GB
# of PyTorch + ~200MB of Ray for a build that doesn't need them.
)
# sdist tarball from PyPI (CDN path is content-addressed and stable per version)
source=("https://files.pythonhosted.org/packages/ec/93/c896156981f56228e6a20ed1a95814aabb75d5e4a97a44534b216a269fdb/skillopt-${pkgver}.tar.gz")
sha256sums=('d9f047336d4d13936f26589677406db77df43897fa23dd7973f81aff7907c4e4')
# ---------------------------------------------------------------------------
# Build: PEP 517 isolated wheel build using the system setuptools backend.
# ---------------------------------------------------------------------------
build() {
cd "${srcdir}/skillopt-${pkgver}"
python -m build --wheel --no-isolation
}
# ---------------------------------------------------------------------------
# python-skillopt: full install of the wheel + CLI entry points + LICENSE.
# ---------------------------------------------------------------------------
package_python-skillopt() {
depends=(
'python-openai>=1.30.0'
'python-yaml>=6.0'
'python-numpy>=1.24.0'
'python-openpyxl>=3.1.0'
'python-azure-identity>=1.15.0'
'python-azure-core>=1.30.0'
'python-httpx>=0.27.0'
)
optdepends=(
'python-alfworld: ALFWorld benchmark environment (skillopt[alfworld] extra)'
'python-gymnasium: ALFWorld backend gym env (skillopt[alfworld] extra)'
'python-omegaconf: ALFWorld config loader (skillopt[alfworld] extra)'
'python-pytorch: ALFWorld deep-learning backend (skillopt[alfworld] extra)'
'python-ray: Parallel rollout execution for ALFWorld (skillopt[alfworld] extra)'
'python-claude-agent-sdk: Claude model backend (skillopt[claude] extra)'
'python-vllm: Qwen local model backend via vLLM (skillopt[qwen] extra)'
'python-json-repair: Robust JSON parsing for Claude/Qwen backends'
'python-datasets: SearchQA data materialization (skillopt[searchqa] extra)'
'mkdocs-material: Build the documentation site (skillopt[docs] extra)'
'mkdocstrings: Build the documentation site (skillopt[docs] extra)'
'python-skillopt-webui: Gradio dashboard meta-package (skillopt[webui] extra)'
'ruff: Linter for development (skillopt[dev] extra)'
'python-pytest: Test runner for development (skillopt[dev] extra)'
# NOTE: skillopt[codex] needs `openai-codex-sdk` from PyPI. It is not yet
# packaged on Arch/AUR, so it cannot be a proper optdepends. Users enabling
# the Codex backend must `pip install --user openai-codex-sdk` (or use a
# venv) — see the post_install() hook in python-skillopt.install.
)
cd "${srcdir}/skillopt-${pkgver}"
python -m installer --destdir="${pkgdir}" dist/*.whl
# License (hardcode the package name; ${pkgname} is an array in split PKGBUILDs)
install -Dm644 LICENSE "${pkgdir}/usr/share/licenses/python-skillopt/LICENSE"
}
# ---------------------------------------------------------------------------
# python-skillopt-webui: meta-package. The WebUI is shipped inside the main
# `skillopt` wheel under the `skillopt_webui` module; installing this split
# only pulls `python-gradio` as a hard runtime dependency so users get the
# dashboard with `pacman -S python-skillopt-webui`.
# ---------------------------------------------------------------------------
package_python-skillopt-webui() {
pkgdesc="Gradio dashboard for python-skillopt (meta-package; pulls python-skillopt + python-gradio)"
depends=(
'python-skillopt'
'python-gradio>=4.0.0'
)
install -dm755 "${pkgdir}/usr/share/doc/python-skillopt-webui"
cat > "${pkgdir}/usr/share/doc/python-skillopt-webui/README.md" <<'EOF'
# python-skillopt-webui
This is an AUR split / meta-package. The WebUI source itself ships inside
the upstream `skillopt` Python wheel (module: `skillopt_webui`). This
package only adds `python-gradio` as a hard runtime dependency so that
`pacman -S python-skillopt-webui` gives you everything needed to launch
the dashboard.
To start the WebUI after installation:
python -m skillopt_webui.app --port 7860
See `python -m skillopt_webui.app --help` for flags (--host, --share, etc.).
EOF
install -Dm644 "${srcdir}/skillopt-${pkgver}/LICENSE" \
"${pkgdir}/usr/share/licenses/python-skillopt-webui/LICENSE"
}
|