Content
73%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 process skill: the workflow is clearly sequenced with explicit stop conditions and evidence requirements, and the guidance is concrete about artifacts and gates. The main drag is conciseness — the essayistic opening and provenance footer spend tokens on motivation rather than instruction.
Suggestions
Cut the motivational opening paragraphs ("Every other check in this chain reads the product... fail the only test that counts") down to one or two lines stating the skill's purpose; the reader is Claude, which does not need persuading.
Drop or compress the provenance footnote at the end — licensing/lineage detail does not help execute the skill and belongs in the referenced LICENSE file itself.
Trim the italicized aside ("This is the same split the rest of the kit uses...") and similar kit-commentary asides; they add context about the framework rather than instructions for the run.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The bulk of the body is dense, non-padded instruction that assumes Claude's competence, but the motivational opening ("Every other check in this chain reads the product. This one uses it... That gap is where the embarrassing defects live") and the closing provenance footnote are unnecessary for execution and could be trimmed. Mostly efficient with some clearly cuttable material. | 3 / 5 |
Actionability | Guidance is concrete and executable for an instruction-only skill: a matrix of (persona, journey, assertion) triples written before launch, a specific report location ("docs/dogfood-reports/, named for the branch and the date"), enumerated ticket fields, and named prerequisites. The actual driving mechanics are delegated to the project's product article, which is a deliberate design choice but leaves minor gaps. | 4 / 5 |
Workflow Clarity | A clear five-step numbered sequence with explicit validation gates throughout: refuse to run on trunk, stop if the suite is red, "wait for it to be genuinely ready rather than merely started", and "If the surface will not come up, that is the finding, and the run stops here". The pre-written matrix acts as a checklist and every failed row requires evidence. The skill performs no destructive or batch operations (it explicitly fixes nothing), so the destructive-operation cap does not apply. | 5 / 5 |
Progressive Disclosure | A single, well-organized file with clear section headers and a logical flow, and no bundle files exist to reference. All content legitimately belongs inline for a process skill of this size, but at ~145 lines with no overview-to-detail split and only pointed-at project files (constitution/local-product.md, /to-tickets), it falls short of the fully-signaled reference structure of the top anchor. | 4 / 5 |
Total | 16 / 20 Passed |