Content
70%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-engineered, gate-driven optimization workflow with excellent sequencing, validation feedback loops, and concrete commands and config. Its two real weaknesses are runtime-blocking: the three bundle files it repeatedly defers to do not exist in the bundle, and the framing sections repeat themselves enough that the body could be materially tightened.
Suggestions
Ship the three referenced bundle files — rig-template.md, test-template.md, and reference/red-test-robustness.md — alongside SKILL.md. The body instructs Claude to 'read it now' at the C-gate and to build the shipped test from test-template.md, so their absence dead-ends the workflow at runtime.
Inline a minimal self-contained instant() test example and the SHELL_MARKER definition in SKILL.md, so the core executable artifact survives even where the templates are unavailable; this also closes the actionability gap where the central deliverable exists only in a missing file.
De-duplicate the framing: 'a ruler, not a stopwatch' and 'the verdict comes from the build, never the platform' each appear twice (invariants section plus Goal/phase A), and phase A restates the invariants section — merge these into one statement to cut roughly 40-60 lines without losing any gate or check.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence (no tutorial padding on Suspense, Playwright, or prerendering) and most content is load-bearing, but framing is noticeably repeated: "a ruler, not a stopwatch" appears in both the invariants section and Goal, "the verdict comes from the build, never the platform" appears in both the invariants section and phase A, and phase A restates much of the invariants section. Two prose-heavy framing sections ("What is invariant, and what is yours", "Reporting to the user") could be tightened or merged. | 3 / 5 |
Actionability | Concrete and executable throughout — `npm ls next @next/playwright`, `npx @next/codemod upgrade`, the `exposeTestingApiInProductionBuild` snippet with per-platform conditions, the `rg` partial-prefetching check, and a blocker-to-guide-pattern table. It falls short of fully copy-paste ready because the central artifact (the shipped `instant()` test and the `SHELL_MARKER` definition it asserts on) is delegated to `test-template.md`, which is not present in the bundle. | 4 / 5 |
Workflow Clarity | Phases P through G are explicitly sequenced, each ends in a gate (C-gate VERIFY-RED, D-gate GREEN-under-lock), and validation is woven in as feedback loops: the untrustworthy-RED two-branch resolution, the revert-to-RED/re-apply-to-GREEN differential, the parity checklist, and base-branch reproduction of pre-existing failures. A explicit whole-workflow stop condition ("the locked test from C is GREEN on the rig, the differential (F) holds, and every item above is checked") closes the loop. | 5 / 5 |
Progressive Disclosure | References are clearly signaled, one level deep, and indexed in a Files section with per-file descriptions — the structure would score 5 if the bundle shipped. But scoring against the actual bundle, all three referenced files (`rig-template.md`, `test-template.md`, `reference/red-test-robustness.md`) are absent from the skill directory, so every "read it now" instruction dead-ends at runtime and the detail layer the skill depends on is inaccessible. | 3 / 5 |
Total | 15 / 20 Passed |