implement-graphql

Implement GraphQL

Use this skill when designing, implementing, or reviewing GraphQL APIs in a Rails application with the graphql-ruby gem.

Core Process

DO NOT proceed to step 3 before step 1 is written and failing.

1. SPEC: Write failing spec (happy path + auth + validation error case) — see TESTING.md. Use AppSchema.execute in spec/graphql/. Never use HTTP controller dispatch for GraphQL specs.

2. TYPE: Define arguments and return types. Use connection_type for pagination shapes. Do not leak internal model names.

3. IMPLEMENT: Create resolver/mutation class delegating to a service object. Use dedicated classes instead of inline field blocks.

4. N+1 CHECK: Use dataloader on every association load. For list resolvers, prime the dataloader with the records returned by the relation before fields resolve associated objects. Use bullet and db-query-matchers in specs.

   # ✅ batches loads across all records
   def buyer
     dataloader.with(Sources::RecordById, Buyer).load(object.buyer_id)
   end
   

5. AUTH CHECK: Apply field-level guards where data is sensitive using Pundit or custom context guards.

   field :internal_notes, String, null: true do
     guard -> (_obj, _args, ctx) { ctx[:current_user]&.admin? }
   end
   

6. FINAL CHECK: Verify every item in the HARD-GATE checklist below. Ensure mutations return { result, errors } shapes on failure.

   rescue ActiveRecord::RecordInvalid => e
     { order: nil, errors: e.record.errors.full_messages }
   

7. RUN: Ensure the full test suite is green before PR.

HARD-GATE Checklist

Before shipping a resolver/mutation slice, ALL of the following must be confirmed:

• Specs — covers happy path, unauthenticated, unauthorized, validation errors, N+1 counts, and schema limits.
• N+1 Prevention — dataloader.with(Source, Model).load(id) on every association; never object.association.
• Dataloader Priming — collection resolvers prime records before association fields resolve.
• Authorization — sensitive fields have field-level guards (not type-level alone).
• Type Conventions — paginated collections use Types::*Type.connection_type, not plain arrays.
• Schema Safeguards — introspection disabled in production; max_depth and max_complexity set.
• Error Handling — mutations return { result, errors } with rescue blocks; no unhandled exceptions.
• Documentation — description: on every field in every type.
• Resolver Structure — dedicated resolver classes, not inline field blocks.