Content
100%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.
An exemplary instruction-only skill body: lean, executable, and precisely sequenced, with a validated fix-and-verify loop, concrete commands and grep hooks for every step, and a documented exception path. Detail is appropriately deferred to DESIGN.md and the design-review rubric.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean throughout: it assumes familiarity with gh, CI, and grep, and every element — the check-reading snippet, the field table, the five-step fix loop, the rules — carries operational information with no padding or explanation of concepts Claude already knows. | 5 / 5 |
Actionability | Guidance is fully executable: the copy-paste-ready gh api snippet for reading notes, the complete reproduce command (`pnpm evals:e2e <slug> --local && pnpm --dir evals design:review -- --test-run latest --json`), a concrete grep mapping (`[data-library-row="…"]` → `data-library-row`), and a sample PR-body line cover the common cases. | 5 / 5 |
Workflow Clarity | The 'Reproduce, fix, verify' section gives a clearly sequenced five-step workflow with an explicit validation checkpoint ("The note must be gone and no new Worth fixing note may appear") and a fix→re-run feedback loop, plus a human-check step ("Look at the screenshot yourself"). | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview that defers detail one level deep with clear signaling — "Read DESIGN.md and the rule in evals/design-review/rubric.md before changing anything" — and the JSON note format for scripts is pointed to rather than restated; nothing that belongs in a separate file is inlined. | 5 / 5 |
Total | 20 / 20 Passed |