---

name: components-and-workflows description: "Compose Superduper components, callable models, listeners, training scaffolds, applications, and lightweight workflow patterns." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Components and Workflows

Use this sub-skill when a task involves composing Superduper AI workflows from components: Component, ObjectModel, Model, Dataset, Metric, Validation, Trainer, Listener, Application, CronJob, FunctionCronJob, or Streamlit.

Route Here For

• Choosing component identifiers and upstream dependencies for multi-step workflows.
• Wrapping Python callables or object methods as ObjectModel instances.
• Calling predict and predict_batches with the key shape expected by a model signature.
• Building listeners that read a table/query key and write model outputs to managed output tables.
• Applying components with db.apply(...), including chained listeners and grouped applications.
• Sketching training, validation, metric, dataset, cron, Streamlit, and simple RAG-like component layouts.
• Diagnosing workflow composition errors around identifiers, callable serialization, listener keys/selects, output naming, and downstream datatype mismatches.

Route Elsewhere

• Nearest-neighbor/vector-index setup, vector search measures, and table.like(...) retrieval belong in vector-search-and-retrieval.
• Plugin/provider installation, credentials, service daemons, and provider-specific optional dependencies belong in plugins-and-integrations.
• Connection strings, artifact stores, config files, metadata stores, and low-level Datalayer setup belong in datalayer-and-config.

Required Reading

1. For component, model, listener, training, validation, and UI/API constructor behavior, read references/components-models-listeners.md.
2. For end-to-end workflow patterns, db.apply(...), listener chaining, Application grouping, and simplified RAG structure, read references/workflows.md.
3. For common failure modes and repair actions, read references/troubleshooting.md.
4. For a safe local construction/prediction smoke helper, use scripts/superduper_component_smoke.py.

Quick Operating Checklist

• Give every component a stable, non-empty identifier; use readable lowercase or snake/kebab names for workflow pieces.
• Represent prerequisites explicitly with upstream=[...], especially for downstream listeners that read upstream listener output fields.
• Prefer ObjectModel(identifier=..., object=top_level_callable, datatype=...) for deterministic local components.
• Match a listener's key shape to the wrapped model signature before applying it.
• Use listener.outputs as the downstream document field name and listener.predict_id when a query API asks for an output id.
• Apply source tables/data before listeners that need a select; use jobs=False only when you want to register components without running jobs.
• Keep vector nearest-neighbor details, provider credentials, and low-level config out of this route.