Invoking Skills¶
The skills run inside Claude Code. Install the plugin bundle once, then trigger skills with plain-English prompts.
1. Install Claude Code¶
If claude --version fails in your terminal, install Claude Code
first via the Claude Code quickstart.
The plugin-management slash commands below are only exposed in the
standalone CLI session — IDE-extension chat panels (VSCode, JetBrains)
do not accept them. Once installed via the CLI, the bundle is loaded
by every surface, IDE extensions included.
2. Install the cuvis-ai skill bundle¶
Open a claude session and run these slash commands:
/plugin marketplace add cubert-hyperspectral/cuvis-ai-agentic-skills
/plugin install cuvis-ai-agentic-skills@cuvis-ai-agentic-skills
/reload-plugins
The eight skills register and become available to subsequent prompts.
3. Trigger a skill¶
Skills auto-trigger on keywords. A few examples:
| Prompt | Skill that fires | Result |
|---|---|---|
| "Set up cuvis-ai on this machine." | cuvis-ai-env |
A verified env plus .cuvis-ai-env.json recording it. |
| "Build me a pipeline that runs RX on cu3s data and exports an anomaly MP4." | cuvis-ai-pipeline → auto-chains to cuvis-ai-pipeline-visualize, then cuvis-ai-inference if you ask to run it. |
A pipeline YAML, a diagram next to it, and the MP4. |
| "Visualize this pipeline yaml." | cuvis-ai-pipeline-visualize |
A <pipeline>.png (or Mermaid .md) next to the YAML. |
"Run rx_statistical.yaml on Demo_000.cu3s." |
cuvis-ai-inference |
The artifact set on disk plus the exact command used. |
| "Design a two-phase training run for this pipeline." | cuvis-ai-training |
A trainrun.yaml plus orchestration script. |
| "Package this fork of YOLO as a cuvis-ai plugin." | cuvis-ai-plugin (delegates to cuvis-ai-node) |
A pip-installable plugin package + manifest entry. |
| "Make this pipeline run in cuvis.next, switch to video mode." | cuvis-ai-pipeline-to-cuvisnext |
A <pipeline>_cuvisnext_video.yaml. |
You can also invoke a skill explicitly with a /<skill-name> slash
command in the Claude Code prompt — for example, /cuvis-ai-pipeline.
Skill state and configuration¶
- Runtime config —
.cuvis-ai-env.jsonlives in the working directory and pins the env path + plugin manifest.cuvis-ai-envwrites it once; the other skills read it before anyuv runinvocation. If you change Python environments, re-runcuvis-ai-env. - Plugin manifest — manifests follow the Plugin Development format.
cuvis-ai-pipelineandcuvis-ai-inferencehonour whatever manifestcuvis-ai-envrecorded. - Working directory — every skill operates relative to the directory you invoked Claude Code from. Run from your project root, not from inside
docs/orexamples/.
Troubleshooting¶
| Symptom | Likely cause | Fix |
|---|---|---|
| Slash command "not found" in IDE | Plugin slash commands are CLI-only | Run the install from a standalone claude session, then return to the IDE. |
| Skill doesn't trigger on a prompt | Keywords didn't match | Invoke explicitly with /<skill-name> or include the trigger keyword from the skills catalog. |
command not found: restore-pipeline mid-run |
Env missing or stale | The running skill auto-invokes cuvis-ai-env. Wait for the env setup to complete and re-prompt. |