SKILL.md
Adding Tests
Core Principle
One test is usually enough. Only add multiple tests if the bug manifests differently under multiple conditions.
Test behavior, not implementation. Validate outcomes, not code paths.
A test's value is measured by:
- Does it fail when user-facing behavior breaks?
- Does it stay green when internals are refactored?
- Does it catch the real issue it was written for?
Routing
Identify your context, then read the corresponding file:
| Context | When to use | Reference |
|---|---|---|
| Bug fix | Fixing a reported issue, adding a regression guard | branches/bug-fix.md |
| New feature | Adding new behavior, validating a spec | branches/new-feature.md |
Universal Rules
These apply regardless of context:
- Prefer integration tests — they catch real bugs; unit tests rarely do for most projects
- Mock as little as possible — use real deps, temp dirs, test guards
- One behavior per test — don't combine multiple assertions for unrelated behaviors
- Test the public surface — never modify production code (add params, expose internals) to make a test work
Read the relevant branch file for context-specific guidance.