---

name: feature-definitions description: "Model Feast entities, fields, data sources, feature views, on-demand/batch/stream transforms, feature services, labels, permissions basics, validation, and versioning without applying a repo." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Feature Definitions

Use this sub-skill when the user asks how to define, migrate, validate, or troubleshoot Feast Python definition objects: Entity, Field, FileSource/request/push sources, FeatureView, OnDemandFeatureView, BatchFeatureView, StreamFeatureView, FeatureService, label-oriented definitions, and permission metadata basics.

Route First

• Definition modeling, constructors, schema/types, ttl, online/offline, vector metadata, source wiring, ODFV decorators, feature services, and validation errors: stay here.
• Feature repo layout, feast apply, feast plan, registry commands, and CLI lifecycle: use ../feature-repos-and-cli/SKILL.md.
• Historical/online retrieval, materialization, push, saved datasets, and point-in-time joins: use ../retrieval-and-materialization/SKILL.md.
• Vector document retrieval or RAG workflow beyond defining vector Fields: use ../rag-and-vector-search/SKILL.md.
• Feature server, remote registry/store, auth enforcement, TLS, and RBAC runtime behavior: use ../servers-and-remote/SKILL.md.

Core References

• API signatures and safe snippets: references/api-reference.md.
• Data-modeling patterns for entities, fields, sources, feature views, services, labels, and permission metadata: references/data-modeling.md.
• Transformations, ODFVs, batch/stream feature views, validation, and versioning: references/transformations.md.
• Diagnostics for imports, schema/type/vector mistakes, source wiring, TTL, and service precompute issues: references/troubleshooting.md.

Safe Definition Validation

Run the bundled validator against a Python definitions file before apply/materialization work:

python skills/disco/feast/sub-skills/feature-definitions/scripts/validate_feature_definitions.py path/to/features.py

Expected output lists discovered Feast object types/names and reports local validation warnings/errors. The script imports the installed feast package if available, never calls FeatureStore.apply, plan, materialize, push, retrieval APIs, or any server command, and does not require a Feast source checkout.