Content
77%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a well-sequenced, highly actionable TDD workflow with explicit validation checkpoints and error-recovery loops. Its weaknesses are redundancy — the GREEN readability primitives and refactor-checklist pointers each appear twice — and a progressive-disclosure structure whose referenced rules files are missing from the bundle.
Suggestions
State the GREEN-phase readability primitives once: either keep the list under 'Code Quality (during GREEN and REFACTOR)' and drop the duplicated prose in Step 2's GREEN section, or vice versa.
Include the referenced bundle files (rules/red.md, rules/green.md, rules/refactor.md, rules/test-provenance.md, rules/test-after.md, lens.md) or remove the pointers — every 'See rules/...' reference currently dangles, so Claude following them finds nothing.
Trim the anti-patterns table entries that restate the Test Quality rules ('Writing tests in bulk' vs 'One test per cycle'; 'Testing private methods' vs 'Test through public API') and keep only the ones that add new guidance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient directive prose, but the GREEN-phase readability primitives are stated twice — "apply the basic readability primitives while you write: meaningful names, guard clauses... no nesting beyond 2 levels" in Step 2 and again as the "GREEN-phase primitives" list under Code Quality — and "Judge the GREEN output against the quality checklist in rules/refactor.md" also appears in both the REFACTOR section and Code Quality. The anti-patterns table further restates "One test per cycle" and "Test through public API". This is more than anchor 4's minor trimming, but there is no padding or teaching of concepts Claude already knows, so it is above anchor 2. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is fully concrete and covers the common cases with specific examples: glob patterns ("**/*.test.*, **/*.spec.*, **/test_*.*"), an example test name ("it('should reject order when inventory is zero')"), a factory pattern ("buildUser(overrides?)"), hard thresholds ("Cognitive complexity ≤ 15", "more than 3 mocks", "10-15 tests per file"), and named refactoring moves ("Extract Guarded Function", "Consolidate Parallel Maps"). Per the scoring notes, absence of code is not penalized when the guidance is this actionable; this clearly fits anchor 5 rather than anchor 4's 'minor gaps'. | 5 / 5 |
Workflow Clarity | Steps 0-4 are clearly sequenced with explicit validation at every checkpoint: "Write exactly ONE failing test. Run it. Confirm it fails with the expected error", "Run the full relevant test suite to check for regressions", and Step 3's "If a test fails, stop and fix it before proceeding. Never accumulate broken tests". The "When Things Go Wrong" section supplies error-recovery feedback loops per phase, matching anchor 5's sequence-plus-validation-plus-feedback-loops pattern rather than anchor 4's 'minor validation gaps'. | 5 / 5 |
Progressive Disclosure | References are clearly signaled and one level deep ("See rules/red.md", "[lens.md](./lens.md)"), but none of the referenced files — rules/red.md, rules/green.md, rules/refactor.md, rules/test-provenance.md, rules/test-after.md, or lens.md — exist in the bundle, so every pointer dangles and the disclosure structure cannot actually be used. Substantial rule content (mocking strategy, quality checklist, anti-patterns) is also inlined. This sits below anchor 4's 'minor organization gaps' but above anchor 2, since the body still functions standalone. | 3 / 5 |
Total | 16 / 20 Passed |