Content
93%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 exceptionally lean, high-signal body: concrete numeric criteria throughout, verified one-level-deep references, and clear findings output structure. The only gap is that the review workflow is implied by section order rather than explicitly sequenced.
Suggestions
Number the review steps explicitly (e.g., 1. Declare surfaces/age band 2. Score each axis 0-3 citing evidence 3. Run the cross-screen consistency check 4. Emit findings rows) so the sequence and its checkpoints are unambiguous.
State what to do when a screen fails a Blocker rule (e.g., stop and report vs. continue scoring) to close the loop on the severity system in the Findings section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~30-line body is all rules and numeric thresholds with no padding and no explanation of concepts Claude already knows; every token earns its place. Not 4 because there is no over-explanation to trim. | 5 / 5 |
Actionability | The guidance is directly checkable: "text contrast 4.5:1 (3:1 large)", "targets 48px parent, 64px child", "parent gate before settings, purchases, external links, and deletion", "destructive actions two-step with export or restore", plus a concrete findings row format ("severity | axis | screen | evidence | consequence | smallest fix | edit prompt"). | 5 / 5 |
Workflow Clarity | The section order (Declare First → Score Six Axes → Findings) with checkpoints like "cite evidence" and the "Cross-screen check" gives a clear sequence with most checkpoints present, but the steps are never explicitly numbered or sequenced as a procedure. Not 5 because explicit validation sequencing (e.g., a numbered review loop) is missing; not 3 because checkpoints are present rather than implicit. | 4 / 5 |
Progressive Disclosure | A short overview with three well-signaled, one-level-deep references ("Criteria: [Rubric](references/rubric.md). Child rules: [Child surface](references/child-surface.md). AI tells: [Human feel](references/human-feel.md)"), all of which exist on disk and contain no nested references. | 5 / 5 |
Total | 19 / 20 Passed |