---

name: packaging-release description: "Operate AnyLabeling install, test, packaging, PyInstaller, translation/resource, CPU/GPU wheel, macOS, and pre-publish release workflows." disable-model-invocation: true metadata: disco-role: operating license: GPL 3.0

Packaging and release

Use this sub-skill when the task is about installing AnyLabeling for development, proving startup compatibility, building distributable artifacts, maintaining Qt resources/translations, or preparing CPU/GPU/macOS releases.

Route here for

• Choosing the correct install variant: default CPU package, Linux/Windows GPU package, macOS CoreML extra, or developer tooling.
• Running the local pre-publish gate: fresh install, unit tests, startup import smoke, optional real-model inference, and multi-Python checks.
• Building wheels/sdists, checking package metadata, or understanding why the GPU package rewrites pyproject.toml to publish as anylabeling-gpu.
• Building PyInstaller executables, macOS folder-mode output, and diagnosing bundled onnxruntime native DLL failures.
• Regenerating Qt translations/resources and checking for the PySide6 toolchain used to rebuild a PyQt6 resource module.

Route elsewhere

• Manual annotation, label JSON, canvas behavior, and dataset export belong to annotation-ui-and-data.
• Auto-labeling model configs, downloads, SAM/YOLO inference, and model registry behavior belong to auto-labeling-models.
• PyPI upload credentials, release approval, or tag creation are outside this skill; this skill explains the checks and build mechanics but does not authorize publishing.

Read first

• references/packaging-and-release.md for install variants, wheel/build behavior, PyInstaller/resource workflows, and platform notes.
• references/pre-publish-checklist.md for the recommended local gate before tagging a release.
• references/troubleshooting.md for dependency drift, headless Qt, macOS PyQt, GPU wheel metadata, PyInstaller ORT DLL, translation tools, and real-inference skips.

Bundled helper

• scripts/check_language_tools.py checks whether the translation/resource toolchain is available without mutating generated files. Run it before following the resource regeneration recipe.

Keep release commands scoped to a disposable or intentionally prepared checkout. Some maintainer scripts mutate package metadata or generated resources; read the linked references before running them.