---

name: runtime-configuration description: "Install, inspect, and configure TotalSegmentator runtime dependencies, devices, weights, licenses, offline caches, and usage statistics." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Runtime Configuration

Use this sub-skill when an agent needs to make TotalSegmentator importable, choose a backend device, inspect local runtime state, pre-stage model weights, configure offline/license behavior, or disable usage telemetry. It owns setup and diagnostics, not segmentation command construction.

Start Here

• Install with pip install TotalSegmentator; verify with TotalSegmentator --version, totalseg_info --json, or the bundled checker.
• Run python scripts/check_totalseg_runtime.py --task total --device cpu --json for a safe smoke check that imports metadata/registry/config helpers, checks PyTorch visibility, and never downloads weights or runs models.
• Read references/install-and-backends.md for Python/PyTorch requirements, optional dependencies, and cpu/gpu/gpu:N/mps selection.
• Read references/configuration.md for TOTALSEG_HOME_DIR, TOTALSEG_WEIGHTS_PATH, config.json, weights, offline staging, licenses, and usage-stat settings.
• Read references/troubleshooting.md when imports, CUDA/MPS, offline weights, licenses, preview rendering, optional dependencies, or config isolation fail.

Routing Boundaries

• For task and class discovery, license flag lookup, or valid --roi_subset names, route to ../capability-discovery/SKILL.md.
• For actual TotalSegmentator -i ... -o ... commands or python_api.totalsegmentator(...) calls, route to ../segmentation-workflows/SKILL.md.
• For run reports, statistics, multilabel outputs, probabilities, or mask combination, route to ../outputs-and-statistics/SKILL.md.
• For DICOM input/output details and crop_to_body, route to ../dicom-and-formats/SKILL.md.
• For phase, modality, body-stats, and Evans-index helper CLIs, route to ../auxiliary-analysis/SKILL.md.
• For training, dataset conversion, or research evaluation workflows, route to ../advanced-training/SKILL.md.

Runtime Rules

• Prefer totalseg_info and totalsegmentator.registry for capability decisions; they are safe, fast, and do not load models.
• Do not log license numbers, absolute local paths, API keys, or machine-specific environment details in shared outputs.
• Treat model downloads as explicit side effects; only run totalseg_download_weights when the user expects network and storage use.
• Use the bundled checker for diagnostics before running expensive segmentation, especially in offline or service environments.