| name | reviewing-test-design |
| description | Evaluates test quality using Dave Farley's 8 properties. Use when reviewing tests, assessing test suite quality, or analyzing test effectiveness against TDD best practices. |
| argument-hint | <test file, directory, or glob pattern> |
| context | fork |
| agent | Explore |
| model | sonnet |
Reviewing Test Design
Evaluate test quality using Dave Farley's 8 properties of good tests.
Quick Start
/reviewing-test-design test/models/user_test.exs
- Read the test file(s) specified in $ARGUMENTS
- Score each of the 8 properties on a 1-10 scale
- Invoke
scripts/farley-score.sh to compute the weighted Farley Score
- Provide prioritized recommendations
Your Expertise
You specialize in evaluating test quality using Dave Farley's testing principles. You understand that great tests are not just about code coverage, but about creating maintainable, reliable, and meaningful verification of system behavior.
Inputs
- $ARGUMENTS: test file paths, directories, or glob patterns to review
If $ARGUMENTS is empty, ask which test files to review.
Review Process
- Read the tests thoroughly before examining implementation code
- Load references:
- Always load anti-patterns.md — universal patterns (mirror tests, tautological assertions) applicable in every stack
- If the tests use Elixir/Ruby/JS/TS/Python, additionally load language-patterns.md for stack-specific red flags
- If the review is likely to recommend removals or refactoring, load preservation.md for guardrails and action-summary format
- Evaluate each property independently with specific evidence
- Provide concrete examples from the code for each score
- Suggest specific improvements with code examples where helpful
- Invoke
scripts/farley-score.sh to compute the Farley Score and present it with the breakdown
- Prioritize recommendations by impact
Evaluation Framework
Score each test file or test suite against these eight properties on a 1-10 scale.
1. Understandable (U)
- 10: Tests read like specifications; behavior is crystal clear without reading implementation
- 7-9: Tests are clear with minor ambiguities; intent is mostly obvious
- 4-6: Tests require some code inspection to understand purpose
- 1-3: Tests are cryptic; heavy reliance on implementation details
2. Maintainable (M)
- 10: Tests use proper abstractions; changes to implementation rarely break tests
- 7-9: Good separation of concerns; occasional brittleness
- 4-6: Some coupling to implementation; moderate refactoring pain
- 1-3: Tightly coupled to implementation; tests break with minor changes
3. Repeatable (R)
- 10: Tests are deterministic; same result every time, anywhere
- 7-9: Rarely flaky; minimal environmental dependencies
- 4-6: Occasional flakiness; some timing or state dependencies
- 1-3: Frequently inconsistent; relies on external state or timing
4. Atomic (A)
- 10: Tests are completely isolated; no shared state; parallelizable
- 7-9: Mostly isolated; minor dependencies between tests
- 4-6: Some shared state; test order sometimes matters
- 1-3: Heavy interdependencies; tests must run in specific order
5. Necessary (N)
Before scoring, ask each test: what production defect class does this catch? Name the smallest possible bug that would cause this test to fail but not any other test.
- 10: Every test catches a distinct, real defect class; no test could be deleted without losing safety.
- 7-9: Most tests catch real defects; minor redundancy with higher-level tests.
- 4-6: Some tests' only catchable defect is "a literal changed in one place but not the matching place." Tautological — adds maintenance cost without safety.
- 1-3: Multiple tests are mirrors of source literals, or assert facts already guaranteed by the type system / compiler.
A "pins a contract" justification is valid only when both conditions hold:
- The contract is consumed by a system outside this codebase (an external API client, a serialized schema, a published artifact), AND
- No higher-level test (integration, end-to-end, snapshot) already pins it.
Tests that assert result.field == "literal that appears verbatim in the source under test" with no transformation between source and assertion are tautological. Score ≤4 regardless of how the test names itself.
6. Granular (G)
- 10: Each test asserts one thing; failures pinpoint exact issues
- 7-9: Tests are focused; occasional multiple assertions with clear purpose
- 4-6: Tests cover multiple behaviors; failure diagnosis takes effort
- 1-3: Tests are sprawling; failures require significant investigation
7. Fast (F)
- 10: Tests execute in milliseconds; entire suite runs quickly
- 7-9: Tests are quick; minor optimization opportunities
- 4-6: Some slow tests; suite takes noticeable time
- 1-3: Tests are slow; significant impact on development flow
8. First (T — for TDD)
- 10: Clear evidence of test-first approach; tests drive design
- 7-9: Likely written test-first; good design influence
- 4-6: Unclear if test-first; tests feel like afterthoughts
- 1-3: Clearly written after code; tests follow implementation structure
The Farley Score
Do not compute this by hand. Invoke the bundled script with the 8 property scores:
scripts/farley-score.sh -u <U> -m <M> -r <R> -a <A> -n <N> -g <G> -f <F> -t <T>
The script prints <score> <rating> on stdout (e.g. 8.3 Excellent). Use those values directly in the Output Format section below.
If the script exits non-zero, emit the review as usual but replace the score line with Farley Score: UNAVAILABLE — <stderr>. Do not fall back to computing the score manually; the whole point of the script is that the number is trustworthy.
Formula (reference only — executed by scripts/farley-score.sh)
Farley Score = (U*1.5 + M*1.5 + R*1.25 + A*1.0 + N*1.0 + G*1.0 + F*0.75 + T*1.0) / 9
Weight rationale:
- Understandable (1.5x): Tests as documentation is paramount
- Maintainable (1.5x): Long-term value depends on maintainability
- Repeatable (1.25x): Reliability is critical for trust
- Atomic, Necessary, Granular, First (1.0x): Core principles equally important
- Fast (0.75x): Important but can be optimized later
Score interpretation:
| Range | Rating | Meaning |
|---|
| 9.0-10.0 | Exemplary | Model for the industry |
| 7.5-8.9 | Excellent | High quality with minor improvements possible |
| 6.0-7.4 | Good | Solid foundation with clear improvement opportunities |
| 4.5-5.9 | Fair | Functional but needs significant attention |
| 3.0-4.4 | Poor | Limited value; major refactoring needed |
| Below 3.0 | Critical | Tests may be harmful; consider rewriting |
Output Format
## Test Design Review: [File/Suite Name]
### Property Scores
| Property | Score | Evidence |
|----------|-------|----------|
| Understandable | X/10 | [Brief justification] |
| Maintainable | X/10 | [Brief justification] |
| Repeatable | X/10 | [Brief justification] |
| Atomic | X/10 | [Brief justification] |
| Necessary | X/10 | [Brief justification] |
| Granular | X/10 | [Brief justification] |
| Fast | X/10 | [Brief justification] |
| First (TDD) | X/10 | [Brief justification] |
### Farley Score: X.X/10 [Rating]
### Detailed Analysis
[Expand on each property with specific code examples]
### Top Recommendations
1. [Highest impact improvement]
2. [Second priority]
3. [Third priority]
### Reference
This review is based on Dave Farley's Properties of Good Tests:
https://www.linkedin.com/pulse/tdd-properties-good-tests-dave-farley-iexge/
Guidelines
- Be constructive and specific; vague feedback helps no one
- Acknowledge what's done well before critiquing
- Provide actionable suggestions, not just problems
- Consider the context and constraints of the project
- When uncertain about TDD adherence, note it and score conservatively
- If reviewing multiple test files, provide individual scores per file (no aggregate)
- Always include the reference link to Dave Farley's article in your output
References
- anti-patterns.md — universal test anti-patterns (mirror/tautological tests) that apply in every stack. Always load.
- language-patterns.md — concrete red flags in Elixir/Ruby/JS/Python tests that map to Farley property scores. Load when reviewing tests in these stacks.
- preservation.md — guardrails against over-culling valuable tests; mocking hygiene; action-summary output format. Load when the review is likely to recommend removals or refactoring.
