fleet-phase-testing
Phase-appropriate testing — what level of rigor for POC vs MVP vs staging vs production. Prevents over-testing early work and under-testing production code.
SKILL.md.claude/skills/fleet-phase-testing/SKILL.md
GitHubPhase-Appropriate Testing
The Problem
Testing a POC with production rigor wastes time. Testing production code with POC rigor creates incidents. The delivery phase determines the testing standard.
Phase Standards
POC (Proof of Concept)
Goal: Does it work at all?
| Test Type | Required | Notes |
|---|---|---|
| Happy path | Yes | Core functionality works |
| Error handling | No | Crashes are acceptable |
| Edge cases | No | Focus on the concept |
| Performance | No | Speed doesn't matter yet |
| Security | Minimal | No auth bypass, no secrets in code |
| Integration | Manual | Run it, see if it works |
| Coverage target | None | Coverage metrics don't apply |
QA focus: Does the concept prove what it set out to prove?
MVP (Minimum Viable Product)
Goal: Can users rely on it?
| Test Type | Required | Notes |
|---|---|---|
| Happy path | Yes | All primary flows work |
| Error handling | Yes | Graceful failures, clear messages |
| Edge cases | Key ones | Common boundary values |
| Performance | Basic | Doesn't hang, responds in reasonable time |
| Security | Standard | Auth works, inputs validated, no injection |
| Integration | Automated | CI runs tests on every push |
| Coverage target | 60%+ | Critical paths covered |
QA focus: Would the PO be comfortable showing this to someone?
Staging
Goal: Production-ready verification
| Test Type | Required | Notes |
|---|---|---|
| Happy path | Yes | Complete coverage |
| Error handling | Yes | All error paths tested |
| Edge cases | Yes | Boundary value analysis complete |
| Performance | Yes | Load testing, response time SLAs |
| Security | Thorough | Penetration testing mindset |
| Integration | Full | End-to-end with real services |
| Coverage target | 80%+ | All modules covered |
QA focus: Does this survive real-world conditions?
Production
Goal: Battle-tested reliability
| Test Type | Required | Notes |
|---|---|---|
| Happy path | Yes | Complete |
| Error handling | Yes | Including cascading failures |
| Edge cases | Yes | Including adversarial inputs |
| Performance | Yes | Under load, under stress, under degradation |
| Security | Complete | Full audit, dependency scanning, secret detection |
| Integration | Full | With monitoring and alerting |
| Coverage target | 90%+ | Including error paths |
| Regression | Yes | Every bug fix gets a regression test |
QA focus: Can we sleep while this runs?
Applying Phase to Fleet Tasks
When predefined test criteria for a task:
- Check the task's
delivery_phasefield - Apply the phase standard from above
- Scale your TC-XXX criteria accordingly
## Test Definition: {task_title}
## Delivery Phase: {phase}
Phase standard applied: {description}
TC-001: ... (happy path — required at all phases)
TC-002: ... (error handling — required at MVP+)
TC-003: ... (boundary value — required at staging+)
Phase Escalation
When a feature moves from one phase to the next, the test suite must be upgraded:
- POC → MVP: Add error handling tests, key boundary values
- MVP → Staging: Add full boundary analysis, performance tests
- Staging → Production: Add adversarial inputs, regression suite
Common Mistakes
- Over-testing POC — Writing 50 tests for throwaway code
- Under-testing MVP — "It works on my machine" is not a test
- Skipping phase upgrade — Feature goes to staging with POC-level tests
- Ignoring the phase field — Testing everything at production level regardless of delivery phase — wastes capacity
- No security at any phase — Even POC should not have hardcoded credentials or SQL injection