Content
81%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 well-crafted procedural skill body: the workflow is explicitly sequenced with validation gates and checklists, and the project-context prelude adds genuinely non-generic information. The main gaps are minor — slight repetition between the prelude and Planning section, and sidecar references that are linked inline rather than presented as labeled navigation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence — the project-context prelude delivers specific, non-generic pointers (file paths, hook names, failure modes) rather than textbook padding. Minor over-explanation remains: the domain-glossary and ADR pointers from the prelude are repeated in the Planning section, and "a test that cannot fail is worse than no test" is stated twice. Not 5 because a few of these spots could be trimmed; not 3 because nothing reads as filler Claude already knows. | 4 / 5 |
Actionability | Concrete, executable guidance throughout: the copy-paste-ready `sh scripts/trace.sh emit ...` command with exact arguments, the WRONG/RIGHT horizontal-vs-vertical diagram, per-cycle checklists, and quoted planning questions. It stops short of 5 because there is no example test or runner command in the file itself — the runner command is deliberately deferred to `constitution/local-engineering.md`, leaving a gap if that file isn't consulted. | 4 / 5 |
Workflow Clarity | The sequence (Planning → Tracer Bullet → Incremental Loop → Refactor) is explicit with validation checkpoints at every stage: RED must fail before GREEN, "Run tests after each refactor step", "Never refactor while RED", and the suite green before and after the refactor-only commit. Checklists for planning, refactoring, and each cycle, plus feedback-loop rules ("Only enough code to pass current test"), match the top anchor. | 5 / 5 |
Progressive Disclosure | The overview is a lean ~130-line procedure with one-level-deep sidecar references (tests.md, mocking.md, deep-modules.md, interface-design.md, refactoring.md) and project-context pointers, so no nested-reference problem exists. It falls below 5 because the links are embedded mid-sentence in checklists rather than clearly labeled navigation sections, and none of the referenced sidecar files are present in the provided bundle to verify they resolve. Not 3: the references are clearly signaled and no content that belongs in a sidecar is inlined. | 4 / 5 |
Total | 17 / 20 Passed |