webrobot-plugin-dev

Implement a stage interface such as WTransformStage or WSinkStage in Scala, register the class in META-INF/services matching the fully-qualified interface name, and declare the stage with its args in manifest.json. Build with Gradle using the webrobot-plugin-sdk from JitPack.

Create a Java class annotated with @Singleton and @Path, implement endpoints with @RequiresScopes, and resolve the tenant via OrganizationContextHelper.getOrganizationId. Package it with Maven, add Flyway migrations under db/migration, and declare the entrypoint in manifest.json.

Use WPartitionStage when the stage needs WebroStageContext access, such as ctx.llm.infer, ctx.query, or ctx.execute for per-row enrichment. WTransformStage is stateless and serialized to Spark workers, so it receives no context.

Stage name collisions at plugin load silently override existing stages. Query the public catalog endpoint at /webrobot/api/catalog/stages before naming a new stage to verify the ID is not already registered.

Every domain table must include organization_id, resolved from the JWT via OrganizationContextHelper.getOrganizationId and never from the request body. All queries must filter by organization_id to isolate tenant data.

Non-Serializable fields like HttpClient must be declared as @transient lazy val so they are re-created after deserialization on each executor. Without this, Spark throws a NotSerializableException when shipping the stage to workers.