Valuable tests
Use when creating, changing, or reviewing maintained automated tests. Optimize for regression value, precision, determinism, and maintenance cost. Skip manual validation and test-infrastructure-only work.From its SKILL.md
npx -y skills add seanforge/skills --skill valuable-testsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.4 KB, as published. Nobody here has run it
Valuable Tests
Maximize confidence per maintenance cost.
Preserve explicit validation requirements and public contracts.
Select tests
- Use risk-based test selection. Each test must protect against a distinct, in-scope regression.
- Stop when an additional test would duplicate an existing failure signal or address only speculative risk.
Choose scope
Choose the narrowest scope that preserves the failure semantics. Add another scope only when it protects a distinct failure class.
- Unit: Verify a local behavioral contract without real collaboration semantics. Test a function directly only when it owns an independent contract; exercise private behavior through that contract.
- Integration: Verify behavior that arises from real collaboration or dependency semantics.
- E2E: Define the SUT explicitly; enter through its supported external interface and traverse its in-scope production path to an observable result. Reserve E2E for critical behavior unavailable at narrower scopes.
Design cases
Treat one representative as sufficient for an equivalence partition. Add boundary values, combinations, or transition sequences only when they change the expected outcome or protect against a selected regression risk.
Assert contracts
- Assert the smallest observable contract that distinguishes the behavior. Multiple assertions are valid only when they jointly define that contract.
- Derive expected results independently; do not reproduce production logic.
- Assert interactions, side effects, and post-failure state only when they are contractual.
Control dependencies
- Use the real implementation for every dependency whose behavior the test claims to verify. Replace only dependencies outside the chosen test boundary. A test double cannot validate the implementation it replaces.
- Do not widen production interfaces or add indirection solely for testing.
Preserve signal
- Tests must be deterministic and order-independent. Control nondeterminism at its source; do not compensate with retries, arbitrary sleeps, widened timeouts, or execution order.
- Make failures identify the violated contract and minimal triggering context.
- Treat coverage metrics and test counts as diagnostics, never targets.