with-testing

Map test boundaries before writing tests — classify each collaborator as real or stubbed, then let risk set depth.

Test Boundary Map

Before writing tests, classify each collaborator of the unit under test:

| Collaborator | Category | Decision | |---|---|---| | HTTP / external API | External | Mock — network is variable, not intention | | Filesystem | External | Mock — side-effect producer | | Date / RNG | Nondeterministic | Mock — breaks reproducibility | | DOM side effects | External | Mock — environment dependency | | Database | External | Context-dependent — prefer test DB when available | | Internal module | Owned | Keep real — mocking what you own verifies nothing |

The heuristic: if a collaborator can fail the test for reasons unrelated to the code's intention, mock it.

For mock implementation rules and TDD workflow, use the tdd skill.

Depth & Coverage

| Risk | Test expectations | |---|---| | Low | Unit coverage on changed logic; one failure case | | Medium | + integration tests at mock seams; perspective table recommended | | High | + security/failure-path scenarios; perspective table required |

Minimum: 90% branch coverage. Target: 100% for security, complex business logic, public APIs. Edge values: 0, min, max, +-1, empty, null.

Anti-Patterns

• Mocking what you own — stubbing internal collaborators produces tests that verify wiring, not behavior
• Vacuous assertions (>= 0, length >= 0)
• Claiming coverage without execution evidence
• Skipping perspective table for medium/high risk changes