using-ui-bundle-salesforce-data

A skill that owns Salesforce data-access patterns inside React UI bundles (projects with a uiBundles/*/src/ directory). It activates for any Salesforce record operation — reading, creating, updating, or deleting — such as building forms that submit to Salesforce, pages that display records, or any code touching standard or custom objects, including when files import from @salesforce/sdk-data or when .graphql files or codegen.yml exist. It explicitly does not cover authentication/OAuth setup, schema changes, the Bulk/Tooling/Metadata APIs, or declarative automation. All data access must go through the Data SDK (@salesforce/sdk-data), which handles authentication, CSRF, and base URL resolution, using sdk.graphql for queries and mutations and sdk.fetch for REST, always with optional chaining because those methods may be undefined in some surfaces.

Before starting, it verifies preconditions: the SDK installed, a schema.graphql at the project root, and custom objects/fields deployed (checked with graphql-search.sh). If preconditions are unmet it may scaffold components, routes, and UI logic with empty data and TODO markers, but does not write GraphQL strings until the schema workflow is complete. Only a fixed set of APIs is permitted — GraphQL via the uiapi namespace, UI API REST, Apex REST, Connect REST, and Einstein LLM — and unsupported paths such as the SOQL query endpoint (blocked at the proxy), Aura-enabled Apex, and the Chatter API are called out with sanctioned alternatives. GraphQL is preferred for record operations, with REST reserved for cases GraphQL cannot cover.

A set of non-negotiable GraphQL rules reflects Salesforce-specific behavior: HTTP 200 does not mean success, so the errors array must always be parsed; the schema is the single source of truth and every name must be confirmed via the search script; @optional belongs on all read-query record fields to survive field-level security; mutations wrap under uiapi(input: { allOrNone }); pagination must always set first: and include pageInfo; and SOQL-derived limits cap subqueries, traversal depth, and records per subquery. The workflow acquires the schema (without ever opening the 265K-line file directly), looks up entities with graphql-search.sh, generates queries and types, and validates with lint and codegen.