Skip to content

Available Skills

The bundle exposes eight skills. Each one has a tight trigger / deliverable contract.

Skill Triggers on Deliverable
cuvis-ai-env "Set up cuvis-ai", "install cuvis-ai", command not found: restore-pipeline, ModuleNotFoundError: cuvis_ai / cuvis, missing ffmpeg / graphviz. Auto-invoked as a prerequisite by every other skill that needs Python. A verified runtime + .cuvis-ai-env.json recording env path, plugin manifest, system-dep status, and smoke-test result.
cuvis-ai-node "Create or extend a cuvis-ai node, define ports, add statistical_initialization or TRAINABLE_BUFFERS, write node tests." A Node subclass plus a passing pytest file.
cuvis-ai-plugin "Package a cuvis-ai plugin, write plugins.yaml, expose nodes via provides:, tag a release." Also: fork an external library (YOLO, SAM3, ByteTrack, …) as a cuvis-ai plugin. A pip-installable plugin package + manifest entry.
cuvis-ai-pipeline "Build a hyperspectral pipeline (cu3s → spectral op → mask / video sink), choose nodes, emit YAML." A pipeline YAML loadable by restore-pipeline / restore-trainrun.
cuvis-ai-pipeline-visualize "Visualize / show / draw / render this pipeline yaml", interactive editor. Auto-invoked by cuvis-ai-pipeline after a YAML is emitted. A <pipeline>.png (or .md Mermaid) next to the YAML, or an interactive cuvis-ai-ui session.
cuvis-ai-training "Design a training run (statistical-only, gradient, two-phase), set execution stages, wire loss / metrics." A trainrun.yaml + Phase 1 / Phase 2 orchestration script.
cuvis-ai-inference "Run / execute this pipeline yaml on <cu3s>", "produce the MP4 / COCO JSON / .npy features". Auto-invoked by cuvis-ai-pipeline after a YAML is emitted. The requested artifact set on disk plus the exact CLI / Python / gRPC / cuvis.next path used.
cuvis-ai-pipeline-to-cuvisnext "Run this pipeline in cuvis.next", "switch to video mode", "switch to cube mode", "convert to cuvis.next". Auto-invoked by cuvis-ai-inference Path 2. A <pipeline>_cuvisnext_<mode>.yaml next to the source, ready for cuvis.next's pipeline-builder UI.

How they cross-reference

  • cuvis-ai-pipeline defers node-level details to cuvis-ai-node and visualization to cuvis-ai-pipeline-visualize, then hands off to cuvis-ai-inference for execution.
  • cuvis-ai-inference Path 2 hands off to cuvis-ai-pipeline-to-cuvisnext when the target is cuvis.next.
  • cuvis-ai-plugin defers node code to cuvis-ai-node.
  • cuvis-ai-training builds on whatever pipeline is already wired.
  • All seven non-env skills delegate runtime setup to cuvis-ai-env. They read ./.cuvis-ai-env.json before running any uv run … command and invoke cuvis-ai-env if it is absent or invalid.

See also

  • About Skills — the model behind skills and how they compose.
  • Invoking Skills — installing the bundle and triggering skills from a prompt.