Content
67%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 body is an unusually well-structured advisory workflow — tight, phased, and concrete — with only minor duplication between its quick-reference tables and the rule files. Its one serious defect is that nearly all of the files it progressively discloses to (the whole rules/ set) are absent from the bundle, so the carefully designed load-on-demand machinery cannot actually be followed.
Suggestions
Ship the 12 rules/*.md files referenced by the body (chart-type-selection, library-selection, accessibility, anti-patterns, plus the eight on-demand files), or remove the references and inline the essential decision tables — as written, Phase 2-4 instruct loading files that do not exist in the bundle.
Trim the Quick reference section's intent→chart and platform→library tables or the duplicated content in the always-on rule files, so the overview does not spend tokens restating what the bundle files own.
Add an explicit verification checkpoint before final output, e.g. confirm the recommended library supports the constraints gathered in Phase 1 (SSR, offline, real-time) and re-check the recommendation against the hard-rules list before emitting the output template.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, tabular, and operational with no padding or explanation of concepts Claude already knows, e.g. the signal-to-file load table and the hard-rules list ("Pie charts: ≤ 5 slices; otherwise use a bar chart"). It stays at 4 rather than 5 because the Quick reference section re-inlines digest tables that duplicate the always-on rule files it points to ("full table in rules/chart-type-selection.md"), and the output-format template is long. | 4 / 5 |
Actionability | Guidance is concrete and executable for an advisory skill: named libraries pinned to numeric brackets ("Web, large datasets (> 10k points), real-time → Apache ECharts (Canvas)"), a fill-in output template, and a signal-driven file-load table. It stays at 4 rather than 5 because fixes are described generically ("Provide a concrete fix (a prop, a code snippet, or a configuration change)") rather than shown, and the actual decision tables and fixes are deferred to rule files. | 4 / 5 |
Workflow Clarity | Six clearly sequenced phases with per-phase inputs and outputs, a batched-questions rule with a confirm-back checkpoint ("If the user already supplied any of these, do not ask again — confirm them back instead"), and a checklist in the output format. It stays at 4 rather than 5 because there are no explicit error-recovery or verification feedback loops (e.g. verifying the recommended library actually supports the required feature before recommending it). | 4 / 5 |
Progressive Disclosure | The on-paper structure is exemplary (always-on files, a signal-driven on-demand load table, one-level-deep references), but scored against the actual bundle, 12 of the 13 referenced paths — the entire rules/ directory, including the files Phase 4 says to "Always load" — do not exist; only references/galleries-and-examples.md is present. The navigation is unresolvable in practice, which lands at 2 rather than 3: the references are clearly signaled, but the disclosed content is missing from the bundle rather than merely poorly organized. | 2 / 5 |
Total | 14 / 20 Passed |