SpatialAI_MCP: Skill for Claude Code

Viash component authoring

A Viash component = a config.vsh.yaml (metadata: arguments, resources, engines)

• a script (Python/R/Bash). Viash compiles it into a CLI, a Docker image, and a Nextflow module.

Steps

1. Identify the stage and its interface. OpenProblems components conform to a comp_* API merged in via __merge__: /src/api/comp_<stage>.yaml. Find the right interface for the pipeline stage you're implementing (see context/ist-preprocessing-pipeline.md) and match its inputs/outputs exactly.

2. Write config.vsh.yaml. Use the modern flat config (no functionality: wrapper) with engines: / runners: (not the legacy platforms:). Follow OpenProblems conventions: merge the stage API, build on the shared base image with a setup partial, include the standard metadata block, and label the nextflow runner with resource directives. This mirrors a real task_ist_preprocessing component:

   __merge__: /src/api/comp_<stage>.yaml
   
   name: my_method
   label: "My Method"
   summary: "One-line biological purpose."
   description: "Longer description with method context."
   links:
     documentation: "https://github.com/openproblems-bio/task_ist_preprocessing"
     repository: "https://github.com/openproblems-bio/task_ist_preprocessing"
   references:
     doi: "10.xxxx/xxxxx"        # the method's paper
   
   arguments:
     - name: "--some_param"
       type: double
       default: 1.0
       description: "Biologically meaningful parameter."
   # Note: --input/--output are usually inherited from the merged comp_<stage> API.
   
   resources:
     - type: python_script
       path: script.py
   
   engines:
     - type: docker
       image: openproblems/base_python:1        # shared base, not raw python:*
       __merge__:
         - /src/base/setup_spatialdata_partial.yaml   # extra deps via setup partial
     - type: native
   
   runners:
     - type: executable
     - type: nextflow
       directives:
         label: [ midtime, lowcpu, lowmem ]     # resource labels