Content
75%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.
A strong, opinionated process skill: clear sequencing, explicit validation gates, useful checklists, and disciplined delegation of detail to sidecars and project files. Its weaknesses are modest — some restateable testing doctrine, executable specifics (runner command, naming convention) living one hop away, no error-recovery branches, and referenced sidecar files that are absent from the bundle as shipped.
Suggestions
Add brief error-recovery guidance to the workflow (e.g., what to do when the tracer-bullet test cannot reach GREEN, or when a refactor step breaks the suite — revert and re-run before continuing).
State the test-runner command and file-naming convention inline (even one line each) so a reader can execute the first RED step without opening constitution/local-engineering.md.
Tighten the Philosophy section's Good/Bad tests paragraphs into a few bullet rules; the horizontal-slicing rationale could likewise compress without losing the directive force.
Ensure the five referenced sidecar files (tests.md, mocking.md, deep-modules.md, interface-design.md, refactoring.md) actually ship alongside SKILL.md — they are linked but missing from the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and directive — the project-context prelude is high-value pointer content, and rules like "Never refactor while RED" and "A test that cannot fail is worse than no test" earn their tokens. Minor over-explanation remains (the Good/Bad tests paragraphs and the horizontal-slicing rationale restate testing doctrine Claude broadly knows), matching the score-4 anchor of efficient content with trimmable instances. It is not 5 because those doctrinal paragraphs could be tightened, and not 3 because there is no library- or concept-level padding. | 4 / 5 |
Actionability | Concrete, executable guidance dominates: explicit RED→GREEN sequences, a planning checklist ("Confirm with user what interface changes are needed"), per-cycle rules ("One test at a time", "Only enough code to pass current test"), and the pointer "scripts/guards.config.sh. Those two regexes are the machine-readable version of 'this change needed a test'". Per the scoring note for instruction-only skills, absent code is not penalized since the guidance is actionable; it stops short of 5 because the most executable details (the runner command, the file-naming convention) are delegated to external files rather than stated, so the reader must hop before executing anything. | 4 / 5 |
Workflow Clarity | The Planning → Tracer Bullet → Incremental Loop → Refactor sequence is clearly laid out with real validation checkpoints: "Write test for first behavior → test fails", "Run tests after each refactor step", "Never refactor while RED", and "a refactor-only pass is its own commit, with the suite green before and after", plus three explicit checklists. It misses 5 because there are no error-recovery feedback loops (e.g., what to do when the tracer-bullet test cannot go green, or when a refactor step breaks the suite); it is well above 3 since checkpoints are explicit, not implicit. | 4 / 5 |
Progressive Disclosure | The file is a well-organized overview with clearly signaled one-level-deep references: "See [tests.md](tests.md) for examples and [mocking.md](mocking.md) for mocking guidelines", plus deep-modules.md, interface-design.md, and refactoring.md linked with stated purposes in the footer. However, none of the five referenced sidecar files exist in the skill's bundle (no references/ directory and no sibling .md files), so the one-level-deep promise cannot be verified — the structure matches score 4's "minor organization gaps" rather than 5's fully navigable split. | 4 / 5 |
Total | 16 / 20 Passed |