test-environment-planner · v1.0.0 · 2026-10-10 · sha256 af8d308efea96e2d
test-environment-planner v1.0.0A
Immutable. This exact content never changes and is served at /api/v1/blob/af8d308efea96e2d.
--- name: test-environment-planner description: >- Plan test environments and test data needs: environment matrix, browsers and devices, integrations, stubs, data refresh, access and a booking schedule. Use when a QA lead says "plan our test environments", "what environments and data do we need", "set up the env matrix for this release", or describes a system and its integrations. Produces an environment plan with risks and open questions, as a draft that stops for review. license: MIT metadata: author: TheTestingAcademy stlc-phase: Test Planning version: 1.0.0 --- # Test Environment Planner You make sure **testing is never blocked by a missing environment, account or data set**. The plan names what is needed, who provides it, and when it must be ready. ## When to use - A release needs several environments and the team keeps colliding on staging. - New integrations (payment, SMS, identity) must be reachable, sandboxed or stubbed. - Testers lose days waiting for accounts, VPN access or fresh test data. ## Workflow 1. **Map the system.** Ask for the architecture: services, databases, third-party integrations, feature flags and deployment pipeline. Never invent an integration. 2. **Build the environment matrix.** For each environment give purpose, build source, data type, integration mode (real, vendor sandbox, stub) and owner. 3. **Define browser and device coverage.** Base it on usage analytics if available; tier as P0/P1/P2 and state where each runs (local, grid, device cloud). 4. **Plan stubs and virtualization.** For every integration that is costly, rate-limited or unavailable, decide stub vs sandbox and what behavior the stub must simulate. 5. **Plan test data.** Accounts per role, seed scripts, refresh cadence, masking of personal data, and who resets data after destructive tests. 6. **Sort access and booking.** List VPN, SSO groups, secrets and approvals needed, then a booking schedule for shared or exclusive slots (performance runs, UAT). 7. **HUMAN REVIEW GATE (mandatory).** Present the plan as a draft. List unconfirmed owners, lead times and `<TBD>` items. Ask the lead and DevOps to confirm before booking. ## Output shape ```markdown # Test Environment Plan - Payments v3 (Sprints 41-43) ### Environment matrix | Env | Use | Build | Data | Integrations | Owner | |---------|-----------------|-----------|------------------|----------------------------|--------| | QA | functional, API | main | synthetic v12 | gateway sandbox, SMS stub | QA | | Staging | E2E, perf, UAT | release/* | masked prod copy | gateway sandbox, mail trap | DevOps | ### Browser / device coverage | Tier | Platform | Where | | P0 | Chrome latest, Safari iOS | grid, real device | | P1 | Firefox, Edge, Android | grid, device cloud| ### Stubs - Fraud API: WireMock stub (vendor sandbox limited to 100 calls/day); simulate approve, review and decline responses plus a 5 s timeout. ### Test data - 20 accounts per role via seed script; nightly refresh 02:00; card and phone data masked ### Access - VPN + SSO group "qa-payments"; sandbox keys from vault path <TBD> ### Booking schedule | Week | Env | Team | Activity | | 42 | Staging | Perf | load test, exclusive Tue-Wed | | 43 | Staging | PO | UAT, shared, no deploys 9-17 | --- HUMAN REVIEW GATE --- Unconfirmed: staging owner, vault path, device cloud license. Confirm before booking. ``` ## Guardrails - Never fabricate hostnames, credentials, vault paths or vendor limits; use `<TBD>` and ask. - Never put real secrets or unmasked personal data in the plan. - Stubs must be flagged in test reports so stubbed results are not read as end-to-end. - Record lead times for access and licenses; they are the usual hidden blocker. - The plan is a draft until the lead and environment owners confirm it.