Content
62%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 workflow architecture is excellent — a gated P→G loop with trustworthy-RED verification, parity, differential testing, and a PR checklist — and the in-body guidance is specific rather than abstract. However, the skill leans on three bundled files that are not present in the bundle, leaving its central executable artifact (the instant() test) unspecified, and the body carries noticeable duplication that inflates token cost.
Suggestions
Ship the referenced bundle files (`rig-template.md`, `test-template.md`, `reference/red-test-robustness.md`) or inline a minimal executable `instant()` test example so the skill's central artifact actually exists for the reader.
Deduplicate: state the rig-agnosticism (verdict comes from the build, not the platform) and the "ruler, not a stopwatch" point once in phase A, and consolidate the three next-partial-prefetching-optimizer handoff mentions into the "After optimization" section.
Tighten the meta sections ("What is invariant, and what is yours", "Reporting to the user") to their decision-relevant core — each currently spends many lines restating points made elsewhere in the body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and free of basic-concept padding, but noticeably repetitive and tightenable: the rig-agnosticism point appears twice nearly verbatim ("a local `next build && next start`, a CI/staging container, and a per-push preview deploy are equally valid rigs" vs. phase A's restatement "the verdict comes from the build, not the platform"), "a ruler, not a stopwatch" is stated twice, and the `next-partial-prefetching-optimizer` handoff appears three times. Meta sections ("What is invariant, and what is yours", "Reporting to the user") could be condensed. Not 4 because the redundancy exceeds minor instances; not 2 because there is no explanation of concepts Claude already knows and the content is high-signal. | 3 / 5 |
Actionability | The body supplies concrete commands (`npm ls next @next/playwright`, `npx @next/codemod upgrade`, the `rg` partialPrefetching check), complete config snippets (`exposeTestingApiInProductionBuild: process.env.EXPOSE_TESTING_API === '1'`), and a blocker-to-pattern decision table — but the executable core is delegated to `test-template.md`, `rig-template.md`, and `reference/red-test-robustness.md`, none of which exist in the bundle, and no `instant()` test example appears inline. That missing key executable detail places it at 3 rather than 4; it is well above 2 given the in-body specificity. | 3 / 5 |
Workflow Clarity | Phases P→G are clearly sequenced with explicit validation gates and feedback loops: "C-gate: do not start optimizing until the RED is verified trustworthy" with a two-branch resolution, "D-gate: phase D is complete when the locked test from phase C passes GREEN under the lock", the parity checklist (E), the revert/re-apply differential (F), a PR checklist, and an explicit stop condition ("the locked test from C is GREEN on the rig, the differential (F) holds, and every item above is checked"). This matches the 5 anchor's checklists, checkpoints, and error-recovery loops. | 5 / 5 |
Progressive Disclosure | In-body structure is good with clearly signaled one-level-deep references and a "Files" section describing each one, but all three referenced files (`rig-template.md`, `test-template.md`, `reference/red-test-robustness.md`) are absent from the bundle (no references/, scripts/, or assets/ directories exist), so the disclosure structure is broken in practice. Not 4 because missing referenced files is a major, not minor, organization gap; not 2 because references are clearly signaled rather than buried and nothing is nested two-plus levels. | 3 / 5 |
Total | 14 / 20 Passed |