QA Engineer

QA Engineer Skill

Protocols

Read protocol: ${PG_PROTOCOLS}/ux-protocol.md (use the read_file tool before continuing).

Read protocol: ${PG_PROTOCOLS}/input-validation.md (use the read_file tool before continuing).

Read protocol: ${PG_PROTOCOLS}/tool-efficiency.md (use the read_file tool before continuing).

Read protocol: ${PG_PROTOCOLS}/visual-identity.md (use the read_file tool before continuing).

Read protocol: ${PG_PROTOCOLS}/freshness-protocol.md (use the read_file tool before continuing).

Read protocol: ${PG_PROTOCOLS}/receipt-protocol.md (use the read_file tool before continuing).

Read protocol: ${PG_PROTOCOLS}/boundary-safety.md (use the read_file tool before continuing).

Read protocol: ${PG_PROTOCOLS}/conflict-resolution.md (use the read_file tool before continuing).

Run shell command before continuing: cat .production-grade.yaml 2>/dev/null || echo "No config — using defaults" (use the execute_shell_command tool).

Run shell command before continuing: cat Claude-Production-Grade-Suite/.orchestrator/codebase-context.md 2>/dev/null || true (use the execute_shell_command tool).

Fallback (if protocols not loaded): Use AskUserQuestion with options (never open-ended), "Chat about this" last, recommended first. Work continuously. Print progress constantly. Validate inputs before starting — classify missing as Critical (stop), Degraded (warn, continue partial), or Optional (skip silently). Use parallel tool calls for independent reads. Use smart_outline before full Read.

Engagement Mode

Run shell command before continuing: cat Claude-Production-Grade-Suite/.orchestrator/settings.md 2>/dev/null || echo "No settings — using Standard" (use the execute_shell_command tool).

Progress Output

Follow Claude-Production-Grade-Suite/.protocols/visual-identity.md. Print structured progress throughout execution.

Skill header (print on start):

━━━ QA Engineer ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Phase progress (print during execution):

  [1/2] Test Planning
    ✓ {N} test cases across {M} categories
    ⧖ building traceability matrix...
    ○ coverage targets

  [2/2] Test Implementation
    ✓ unit: {N} tests
    ✓ integration: {N} tests
    ⧖ e2e: writing user flow specs...
    ○ performance: load tests

Completion summary (print on finish — MUST include concrete numbers):

✓ QA Engineer    {N} tests written, {M} passing, {K} failing    ⏱ Xm Ys

Brownfield Awareness

If Claude-Production-Grade-Suite/.orchestrator/codebase-context.md exists and mode is brownfield:

• READ existing tests first — understand test framework, patterns, fixtures, helpers
• MATCH existing test framework — if they use pytest, don't introduce jest. If they use Vitest, use Vitest
• ADD tests alongside existing ones — don't restructure their test directory
• Existing tests must still pass — run the full test suite after adding new tests
• Reuse existing fixtures and helpers — don't duplicate test utilities

Config Paths

Read .production-grade.yaml at startup. Use these overrides if defined:

• paths.services — default: services/
• paths.frontend — default: frontend/
• paths.tests — default: tests/

Context & Position in Pipeline

This skill runs AFTER the Software Engineer and Frontend Engineer skills have completed. It expects:

• services/ and libs/ — Backend services, handlers, repositories, domain models, API route definitions
• frontend/ — UI components, pages, hooks, state management, API client calls
• api/, schemas/, docs/architecture/ — API contracts (OpenAPI/AsyncAPI specs), data models, sequence diagrams
• BRD or PRD — Acceptance criteria, user stories, business rules, edge cases

The QA Engineer does NOT modify source code. It generates test files and test infrastructure to tests/ at the project root, and test documentation (test plan, reports) to Claude-Production-Grade-Suite/qa-engineer/.

Graceful Degradation

At startup, check whether frontend/ (or paths.frontend from config) exists. If the frontend directory is not found:

• Skip all frontend-related test phases (UI E2E, visual regression, frontend contract tests, frontend-specific checks).
• Print: [DEGRADED: frontend not found — skipping frontend tests]
• Continue with all backend test phases normally.

Output Structure

This skill produces output in two locations: test deliverables (code, configs, fixtures) at tests/ in the project root, and workspace artifacts (test plan, reports, findings) in Claude-Production-Grade-Suite/qa-engineer/. Never write test files into services/ or frontend/ directly.

Project Root Output (tests/)

tests/
├── unit/
│   └── <service>/                      # One folder per backend service
│       ├── handlers/
│       │   └── <handler>.test.ts       # HTTP handler / controller tests
│       ├── services/
│       │   └── <service>.test.ts       # Business logic / domain service tests
│       ├── repositories/
│       │   └── <repo>.test.ts          # Data access layer tests (mocked DB)
│       ├── validators/
│       │   └── <validator>.test.ts     # Input validation tests
│       └── mappers/
│           └── <mapper>.test.ts        # DTO / domain mapper tests
├── integration/
│   ├── docker-compose.test.yml         # Test dependency containers (Postgres, Redis, Kafka, etc.)
│   ├── setup.ts                        # Global integration test setup / teardown
│   └── <service>/
│       ├── db/
│       │   └── <repo>.integration.ts   # Real DB queries via testcontainers
│       ├── cache/
│       │   └── <cache>.integration.ts  # Real Redis / cache operations
│       ├── messaging/
│       │   └── <queue>.integration.ts  # Real message broker publish / consume
│       └── api/
│           └── <endpoint>.integration.ts  # HTTP-level integration (supertest / httptest)
├── contract/
│   ├── pacts/
│   │   ├── consumer/
│   │   │   └── <consumer>-<provider>.pact.ts  # Consumer-driven contract tests
│   │   └── provider/
│   │       └── <provider>.verify.ts           # Provider verification tests
│   ├── schema/
│   │   └── <api>.schema.test.ts               # OpenAPI schema validation tests
│   └── pact-broker.config.ts                  # Pact Broker connection config
├── e2e/
│   ├── api/
│   │   ├── flows/
│   │   │   └── <user-flow>.e2e.ts     # Multi-step API workflow tests
│   │   ├── smoke.e2e.ts               # Critical-path smoke tests
│   │   └── setup.ts                   # API E2E auth helpers, base URLs
│   └── ui/
│       ├── pages/                     # Page Object Models
│       │   └── <page>.page.ts
│       ├── flows/
│       │   └── <user-flow>.spec.ts    # Playwright / Cypress user flow specs
│       ├── visual/
│       │   └── <component>.visual.ts  # Visual regression snapshot tests
│       └── playwright.config.ts       # Or cypress.config.ts
├── performance/
│   ├── load-tests/
│   │   └── <scenario>.k6.js           # k6 load test scripts (sustained load)
│   ├── stress-tests/
│   │   └── <scenario>.k6.js           # k6 stress test scripts (breaking point)
│   ├── spike-tests/
│   │   └── <scenario>.k6.js           # k6 spike test scripts (sudden burst)
│   ├── baselines/
│   │   └── <scenario>.baseline.json   # Expected p50/p95/p99 latency, throughput
│   └── thresholds.js                  # Shared k6 threshold definitions
├── fixtures/
│   ├── factories/
│   │   └── <entity>.factory.ts        # Test data factories (fishery / factory-girl pattern)
│   ├── seed-data/
│   │   ├── <entity>.seed.json         # Static seed data for integration / E2E
│   │   └── seed-runner.ts             # Script to load seed data into test DBs
│   └── mocks/
│       ├── <external-api>.mock.ts     # External API mock servers (MSW / nock)
│       └── <service>.stub.ts          # Internal service stubs
└── coverage/
    └── thresholds.json                # Per-service and global coverage gates

Workspace Output (Claude-Production-Grade-Suite/qa-engineer/)

Claude-Production-Grade-Suite/qa-engineer/
├── test-plan.md                        # Master test plan with traceability matrix
├── coverage-report.md                  # Coverage analysis and findings
└── findings.md                         # QA findings and recommendations

Phases

Execute each phase sequentially. Do NOT skip phases. Each phase builds on the outputs of the previous one.

Parallel Execution Strategy

After Phase 1 (Test Planning), Phases 2-6 run in parallel — each test type is independent:

# After test plan is written, spawn all test types simultaneously:
<!-- v0.1: do this work yourself; no subagent spawn --> Agent(prompt="Write unit tests following Phase 2 rules. Read test-plan.md for traceability. Write to tests/unit/.", ...)
<!-- v0.1: do this work yourself; no subagent spawn --> Agent(prompt="Write integration tests following Phase 3 rules. Read test-plan.md. Write to tests/integration/.", ...)
<!-- v0.1: do this work yourself; no subagent spawn --> Agent(prompt="Write contract tests following Phase 4 rules. Read test-plan.md. Write to tests/contract/.", ...)
<!-- v0.1: do this work yourself; no subagent spawn --> Agent(prompt="Write E2E tests following Phase 5 rules. Read test-plan.md. Write to tests/e2e/.", ...)
<!-- v0.1: do this work yourself; no subagent spawn --> Agent(prompt="Write performance tests following Phase 6 rules. Read test-plan.md. Write to tests/performance/.", ...)

Wait for all 5 agents to complete, then run Phase 7 (Test Infrastructure) sequentially — it needs all test files to configure CI.

Why this works: Each test type reads source code independently and writes to its own directory. No conflicts. The test plan from Phase 1 provides shared context.

Execution order:

1. Phase 1: Test Planning (sequential — foundational)
2. Phases 2-6: Unit + Integration + Contract + E2E + Performance (PARALLEL)
3. Phase 7: Test Infrastructure (sequential — needs all test files)

Phase 1 — Test Planning

Goal: Produce a traceability matrix linking every BRD acceptance criterion to concrete test cases, categorized by test type.

Inputs to read:

• BRD / PRD acceptance criteria (every GIVEN/WHEN/THEN or equivalent)
• api/ API contracts (OpenAPI specs, AsyncAPI specs)
• schemas/ data models and docs/architecture/ sequence diagrams
• services/ service structure (list all services, handlers, repos)
• frontend/ component and page structure (if frontend exists; otherwise skip frontend inputs)

Actions:

1. Extract every acceptance criterion and assign a unique ID (AC-001, AC-002, ...).
2. For each criterion, determine which test types are required (unit, integration, contract, e2e, performance).
3. Identify all services, modules, and components that need test coverage.
4. Identify all external dependencies that require mocking or test containers.
5. Identify critical user flows for E2E coverage.
6. Identify performance-sensitive endpoints for load testing.
7. Define coverage thresholds per service (lines, branches, functions).

Output: Write Claude-Production-Grade-Suite/qa-engineer/test-plan.md with the following sections:

• Scope — What is being tested, what is explicitly out of scope
• Test Strategy — Test pyramid approach, which test types cover which risk areas
• Traceability Matrix — Table mapping AC-ID to test case IDs, test type, and priority
• Environment Requirements — Containers, external services, env vars needed
• Coverage Targets — Per-service and global coverage gates
• Risk Register — Areas with high complexity or insufficient testability

Phase 2 — Unit Tests

Goal: Test each service's business logic, handlers, and repositories in isolation with full mocking of external dependencies.

Inputs to read:

• services/ source code for each service
• The test plan from Phase 1

Rules:

1. One test file per source file. Mirror the source directory structure under tests/unit/<service>/.
2. Mock ALL external dependencies: databases, caches, message brokers, HTTP clients, other services.
3. Use dependency injection or module mocking — never patch globals.
4. Test the happy path, error paths, edge cases, and boundary values for every public function.
5. For handlers/controllers: test request parsing, validation error responses, correct status codes, response body shape.
6. For services/domain logic: test business rule enforcement, state transitions, calculation correctness.
7. For repositories: test query construction, parameter binding, result mapping (with mocked DB driver).
8. For validators: test every validation rule, including null, empty, boundary, and malformed inputs.
9. Every test must have a descriptive name that reads as a specification: it("should return 404 when order does not exist for the given user").
10. Use factories from tests/fixtures/factories/ for test data — never inline large object literals.
11. Assert on specific values, not just truthiness. Prefer toEqual over toBeTruthy.
12. Test error types and messages, not just that an error was thrown.

Output: Write test files to tests/unit/<service>/.

Also write factories to tests/fixtures/factories/ as you discover entity shapes.

Phase 3 — Integration Tests

Goal: Test service interactions with real dependencies using testcontainers or docker-compose.

Inputs to read:

• services/ database migrations, schemas, connection configs
• docs/architecture/ infrastructure requirements (which DBs, caches, brokers)
• The test plan from Phase 1

Rules:

1. Write tests/integration/docker-compose.test.yml with containers for every real dependency (PostgreSQL, Redis, Kafka, Elasticsearch, etc.). Pin exact image versions.
2. Write tests/integration/setup.ts with global before/after hooks: start containers, run migrations, seed base data, tear down after suite.
3. Each integration test file connects to real containers — no mocks for the dependency under test.
4. Test actual SQL queries against a real database with realistic data volumes (not just 1 row).
5. Test cache read/write/eviction with a real Redis instance.
6. Test message publishing and consumption with a real broker.
7. Test API endpoints with real HTTP calls (supertest / httptest) against a running server.
8. Each test must clean up its own data. Use transactions with rollback, or truncate tables in afterEach.
9. Tests must be parallelizable — use unique identifiers to avoid cross-test data collisions.
10. Test failure modes: connection timeouts, constraint violations, concurrent writes, deadlocks.

Output: Write test files to tests/integration/<service>/.

Write docker-compose.test.yml and setup.ts to tests/integration/.

Phase 4 — Contract Tests

Goal: Verify API consumers and providers agree on request/response schemas and that implementations conform to OpenAPI specifications.

Inputs to read:

• api/ OpenAPI specs and AsyncAPI specs
• services/ API route definitions, request/response DTOs
• frontend/ API client calls and expected response shapes (if frontend exists; otherwise skip consumer-side frontend contracts)

Rules:

1. For each API consumer (frontend, other services), write a Pact consumer test that defines the expected interactions.
2. For each API provider, write a Pact provider verification test that replays consumer expectations against the real provider.
3. Write schema validation tests that load the OpenAPI spec and validate every endpoint's actual response against the schema.
4. Test backward compatibility: if there are versioned APIs, verify old consumers still work with new providers.
5. For async APIs (events, messages), write contract tests for message schemas using AsyncAPI specs.
6. Configure Pact Broker connection in pact-broker.config.ts (even if the broker URL is a placeholder).
7. Contract tests must fail if a required field is removed, a type changes, or a new required field is added without consumer agreement.

Output: Write contract tests to tests/contract/.

Phase 5 — E2E Tests

Goal: Test critical user flows end-to-end through the full stack.

Inputs to read:

• BRD / PRD user stories and acceptance criteria (especially the critical path)
• frontend/ pages and navigation flow (if frontend exists; otherwise API-only E2E)
• services/ API endpoints
• The test plan from Phase 1 (critical user flows identified)

Rules:

1. Identify the 5-10 most critical user flows (signup, login, core CRUD, payment, etc.).
2. For API E2E: chain multiple API calls that represent a complete user journey. Use real auth tokens. Validate side effects (DB state, emails sent, events published).
3. For UI E2E (skip if frontend not found): use Page Object Model pattern. Each page gets a class in tests/e2e/ui/pages/.
4. UI tests must use resilient selectors: data-testid attributes, ARIA roles — never CSS classes or DOM structure.
5. Write a smoke test suite (smoke.e2e.ts) that covers the absolute minimum "is the app alive" checks. This runs on every deploy.
6. E2E tests must be idempotent — running them twice produces the same result.
7. Include setup/teardown that creates test users, seeds required data, and cleans up after.
8. Add explicit waits for async operations — never use arbitrary sleep() calls.
9. For visual regression (skip if frontend not found): capture screenshots of key pages and compare against baselines.
10. Configure test timeouts generously (30s+ per test) — E2E is slow by nature.
11. Cross-boundary journey testing (boundary-safety protocol pattern 5): For every multi-system flow (auth, payment, email, webhook), write at least one E2E test that traces the COMPLETE journey from user action to final state. Auth test must verify: unauthenticated user visits protected page → redirected to login → authenticates → redirected back to original page → sees authenticated content. Payment test must verify: user clicks pay → payment provider processes → callback fires → order status updates → user sees confirmation. Do NOT just test individual hops — test the full chain.
12. Framework navigation correctness: Verify that no <Link> or client-side navigate() targets API routes, external URLs, or auth endpoints. These must use raw <a href> or window.location for full HTTP requests.

Output: Write E2E tests and page objects to tests/e2e/. Write Playwright or Cypress config.

Phase 6 — Performance Tests

Goal: Establish performance baselines and create load/stress test scripts for performance-sensitive endpoints.

Inputs to read:

• docs/architecture/ NFRs (latency targets, throughput requirements, SLOs)
• services/ API endpoints (especially high-traffic ones)
• The test plan from Phase 1 (performance-sensitive areas)

Rules:

1. Write k6 scripts (JavaScript). Each script targets a specific scenario (e.g., "user browsing products", "checkout flow under load").
2. Load tests: simulate sustained normal traffic. Define realistic ramp-up patterns (e.g., 0 -> 100 VUs over 2 min, hold 10 min, ramp down).
3. Stress tests: find the breaking point. Ramp VUs aggressively until error rate exceeds 5% or p99 exceeds SLO.
4. Spike tests: simulate sudden traffic bursts (0 -> 500 VUs in 10 seconds).
5. Define thresholds in each script: http_req_duration['p(95)'] < 500, http_req_failed < 0.01.
6. Write baseline JSON files that record expected performance under normal load. CI compares against these.
7. Use realistic test data — not the same request repeated. Parameterize with CSV data files or k6 SharedArray.
8. Include authentication in test scripts (token generation, session management).
9. Test both read-heavy and write-heavy endpoints separately.
10. Add custom metrics for business-critical operations (e.g., order_processing_time).

Output: Write k6 scripts to tests/performance/. Write baseline files to tests/performance/baselines/.

Phase 7 — Test Infrastructure

Goal: Configure CI test execution, coverage enforcement, and test reliability tooling.

Inputs to read:

• All test files generated in Phases 2-6
• Coverage thresholds from the test plan
• Project CI/CD system (GitHub Actions, GitLab CI, etc.)

Actions:

1. Write tests/coverage/thresholds.json with per-service and global coverage gates:

   {
     "global": { "lines": 80, "branches": 75, "functions": 80, "statements": 80 },
     "services": {
       "<service-name>": { "lines": 85, "branches": 80, "functions": 85, "statements": 85 }
     }
   }
   

2. Write .github/workflows/test.yml (or ci/test-config.yml) with:
   • Unit test stage — runs first, fast, no containers. Fails fast on coverage threshold breach.
   • Integration test stage — starts docker-compose dependencies, runs integration suite, tears down.
   • Contract test stage — runs Pact tests, publishes results to broker.
   • E2E test stage — deploys to test environment, runs smoke + full E2E suite.
   • Performance test stage — runs load tests against staging, compares to baselines.
   • Parallel execution: split unit and integration tests across multiple CI runners by service.
   • Test result artifacts: JUnit XML reports, coverage HTML reports, k6 JSON results.
   • Flaky test detection: track test pass/fail history, quarantine tests with >5% flake rate.
   • Retry policy: retry failed E2E tests up to 2 times before marking as failed.
3. Write seed data runner to tests/fixtures/seed-data/seed-runner.ts.
4. Write external API mock configurations to tests/fixtures/mocks/.

Output: Write CI config to .github/workflows/test.yml, coverage thresholds and test infrastructure to tests/.

Common Mistakes

Execution Checklist

Before marking the skill as complete, verify:

• Claude-Production-Grade-Suite/qa-engineer/test-plan.md has a traceability matrix covering every BRD acceptance criterion
• Every service in services/ has corresponding unit tests in tests/unit/
• Every repository/data-access module has integration tests with real database containers
• Every API endpoint has at least one contract test validating its schema
• The top 5-10 critical user flows have E2E tests
• At least 3 performance-sensitive endpoints have k6 load test scripts with baselines
• tests/integration/docker-compose.test.yml defines all required test containers with pinned versions
• tests/coverage/thresholds.json defines realistic per-service coverage gates
• .github/workflows/test.yml orchestrates all test stages with parallelization and artifact collection
• All test factories are in tests/fixtures/factories/ and reused across test types
• No test file has hardcoded secrets, credentials, or environment-specific values
• All tests can run independently and in any order

This skill body has been adapted for QwenPaw. Differences vs the upstream Claude Code plugin to be aware of:

• No AskUserQuestion tool. When this skill says to surface a decision, render numbered options as plain Markdown and ask the user to type the option name. Parse free-text replies leniently.
• No Skill tool. Phase transitions happen in-line: read the next sub-skill body via read_file from the workspace skills/ dir.
• No subagent spawn. v0.1 is a single-agent flow. If the methodology says "delegate to specialist X", invoke X by reading its SKILL.md from skills/<name>/SKILL.md and following its instructions yourself.
• No TaskCreate/TaskList. Track progress by writing receipts to Claude-Production-Grade-Suite/.orchestrator/receipts/<task>-<role>.json and emitting a one-line status update in chat after each phase.
• WebSearch is tavily_search. Requires TAVILY_API_KEY. If unset, skip the Freshness Protocol and note it.