Contract Testing

<objective> A provider renames a response field. Both services pass their own unit tests, and the mismatch only surfaces in production when the consumer's frontend breaks. Contract testing catches that in CI: the consumer declares exactly what it needs, the provider verifies it can deliver, and can-i-deploy blocks the deploy until the broker confirms both sides are compatible. This skill produces Pact consumer tests, broker-driven provider verification, and the deployment gate that ties them together — so services can be deployed independently without a shared integration environment. </objective>

Discovery Questions

Check .agents/qa-project-context.md first — if it exists, use it and skip anything already answered there. Then:

1. Architecture: Microservices, monolith with separate consumers (mobile/SPA), or BFF pattern? Contract testing matters most when teams deploy independently.
2. Who owns the contract? Consumer-driven (consumers define what they need) or provider-driven (provider publishes a spec)? Most teams benefit from consumer-driven.
3. API versioning strategy: URL-based (/v1/, /v2/), header-based, or none? Contracts must account for version negotiation.
4. How many consumer-provider pairs? Start with the highest-traffic or most-fragile integration. Do not try to contract-test everything at once.
5. Existing API specs: Is there an OpenAPI/Swagger spec? If yes, consider schema-first contracts (or Schemathesis) as a starting point.
6. HTTP or async? Request/response APIs use the HTTP examples here; queue/topic integrations (Kafka, SNS/SQS) use Pact message contracts (see Pact-JS Setup).

Core Principles

1. Consumers define what they need, providers verify they can deliver. The consumer writes a test declaring "I will call GET /users/123 and expect { id, name, email }." The provider runs this test against its real implementation. If the provider cannot satisfy the contract, the build breaks before deployment.

2. The broker is the shared source of truth. Not documentation, not Slack threads, not "just deploy and see." Pacts and verification results live in the Pact Broker; provider verification pulls pacts from the broker (pactBrokerUrl + consumerVersionSelectors), not from local files, because the broker knows which consumer versions are actually live.

3. Contract tests replace integration environments, not integration tests. You still need integration tests for complex multi-step workflows. Contract tests eliminate the need to deploy consumer and provider together just to verify the interface.

4. Break the build on contract violation. A contract test that logs a warning but allows deployment provides zero value. Contracts must be deployment gates.

5. Test the contract, not the business logic. Consumer tests verify response shape and status codes. Provider verification ensures the contract is satisfiable. Business rules belong in unit and integration tests.

Pact-JS Setup

Install @pact-foundation/pact as a dev dependency on both the consumer and provider sides. The workflow has two halves:

• Consumer test: the consumer declares what it needs from the provider (request shape + expected response). Running the test generates pacts/<consumer>-<provider>.json — the contract file. Use Matchers (Matchers.like, Matchers.eachLike, Matchers.integer, Matchers.string, Matchers.regex) so contracts assert types and formats, not brittle exact values.
• Provider verification: the provider runs the consumer's pact against its real implementation (not mocks), using stateHandlers to set up the data each given(...) state expects, and publishes the verification result back to the broker.

Pact-JS v16 (current as of June 2026) renamed `PactV4` → `Pact` and `MatchersV3` → `Matchers`. The old names were removed in v16. If you copy from older blog posts/examples, update the imports. The API behavior is unchanged.

For event-driven systems, the same Pact class supports message pacts (Kafka, SNS/SQS, RabbitMQ) — the consumer asserts the shape of a message it expects and the provider verifies what its producer emits.

See references/pact-js-setup.md for the install commands, the consumer test (single user, 404, paginated list), the broker-driven provider verification spec with state handlers and pending pacts, the Pact Broker Docker Compose, and the message-contract pointer.

Pact Broker

The Pact Broker is the central registry where pact files are published and provider verification results are recorded. It enables the can-i-deploy workflow. Run it locally with Docker Compose backed by Postgres; consumer CI publishes pacts to it tagged with a commit SHA and branch.

Inject every credential from the environment — Postgres password, broker DB URL, and basic-auth password. Hardcoding any of them in the compose file leaks secrets into version control. Pin the broker image to a released tag, not :latest.

See references/pact-js-setup.md for the docker-compose.pact-broker.yml and the pact-broker publish command.

Consumer-Driven Workflow

The full cycle:

1. Consumer writes contract test
   └── Generates pact JSON file

2. Consumer CI publishes pact to broker
   └── Broker stores pact tagged with consumer version + branch

3. Broker webhook triggers provider verification
   └── Provider CI pulls latest pact, runs verification

4. Provider publishes verification result to broker
   └── Broker records: "provider v2.3.1 satisfies consumer v1.5.0"

5. Before deploy: can-i-deploy check
   └── "Can consumer v1.5.0 be deployed? Yes, provider v2.3.1 is in production and verified."

Both pipelines run contract tests, publish results to the broker, and gate deployment on can-i-deploy. The provider pipeline also listens for a repository_dispatch event so a new pact triggers verification automatically.

See references/ci-pipelines.md for the consumer CI workflow, the provider CI workflow (with Postgres service + migrations), and the standalone can-i-deploy / record-deployment commands.

Pact Broker Webhooks

Configure webhooks in the Pact Broker to trigger provider verification via repository_dispatch when a new pact is published. The webhook sends a POST to https://api.github.com/repos/myorg/user-service/dispatches with event type pact-changed, which the provider pipeline listens for (see the repository_dispatch trigger in references/ci-pipelines.md).

Pending Pacts (Incremental Adoption)

Set enablePending: true (plus includeWipPactsSince) on the provider Verifier so a brand-new consumer interaction can land without breaking the provider build — it is reported but does not fail until the consumer marks it expected. This is the standard safety net when adding contracts incrementally.

Schema-First vs Consumer-First

Consumer-First (Pact)

Consumers define what they need; contracts emerge from real usage patterns.

Best for: Teams where consumers have specific needs that differ across clients (mobile needs fewer fields than web), APIs that evolve organically, microservice ecosystems.

Schema-First (OpenAPI + Validation)

Provider publishes an OpenAPI spec; consumers validate their usage against the spec.

Best for: Public APIs with many consumers, APIs designed upfront before implementation, teams with strong API design governance.

OpenAPI 3.0 is not plain JSON Schema (nullable: true etc.) — vanilla Ajv defaults to draft 2020-12 and mis-validates real 3.0 specs. Configure Ajv for the OpenAPI dialect with ajv-formats, or use an OpenAPI-aware validator. See the caveat in references/schema-first.md for the Ajv config and the OpenAPI-against-spec validation helper.

Hybrid Approach

Use OpenAPI as the design artifact and Pact as the enforcement mechanism.

1. Design API with OpenAPI spec (provider team leads design).