---

name: cli-and-configuration description: "Use the TIAToolbox console safely: discover commands, choose the right workflow owner, quote JSON/YAML configuration, validate installs/backends, and troubleshoot CLI startup issues." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

TIAToolbox CLI and Configuration

Use this sub-skill when a user starts from the tiatoolbox console, needs command discovery, asks how to build a shell-safe command, or reports CLI/configuration failures. TIAToolbox exposes the console entry point tiatoolbox=tiatoolbox.cli:main; start with tiatoolbox --help and tiatoolbox --version for safe probes.

Route the Request

• Use references/cli-reference.md to map top-level commands and representative flags before constructing commands.
• Use references/configuration.md for JSON-list/dict quoting, YAML IO config behavior, device strings, and output naming patterns.
• Use references/troubleshooting.md when tiatoolbox is missing, imports fail, native libraries/codecs are unavailable, models cannot download, output paths conflict, or visualization ports collide.
• Use scripts/cli_help_probe.py for safe timeout-bounded help/version checks that do not need a source checkout.

Command Ownership

• Route slide-info, slide-thumbnail, save-tiles, and read-bounds to wsi-io for WSI metadata, reading, tiling, and resolution semantics.
• Route tissue-mask and stain-norm to image-preprocessing for tissue masking, stain normalization, target images, and preprocessing quality decisions.
• Route patch-predictor, semantic-segmentor, nucleus-instance-segment, nucleus-detector, multitask-segmentor, and deep-feature-extractor to model-inference for model selection, pretrained weights, IO config semantics, inference output interpretation, and performance tuning.
• Route visualize and show-wsi to annotation-visualization for TileServer/Bokeh usage, annotation overlays, color maps, and browser/server behavior.

Safe CLI Workflow

1. Probe availability with python scripts/cli_help_probe.py --commands semantic-segmentor --version or direct tiatoolbox --help.
2. Build commands as argument lists first; shell-quote only at the final boundary with shlex.quote or equivalent.
3. Quote JSON arguments as one shell token, especially --input-resolutions, --output-resolutions, and --class-dict.
4. Prefer explicit --device cpu for portable examples; use cuda or mps only after checking the installed torch backend.
5. Choose a new --output-path for model commands unless intentionally passing --overwrite True, because model preparation rejects an existing output directory.

What Not To Duplicate Here

This sub-skill should stay CLI-centric. Deep WSI coordinate rules, image preprocessing theory, model output schemas, annotation store usage, and visualization API details belong in their routed workflow sub-skills.