Content
78%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-structured, highly actionable advisory workflow with a clear phase sequence, concrete decision tables, and a disciplined hand-off story for sibling skills. Its two real weaknesses are broken progressive disclosure — almost every `rules/*.md` file the body instructs Claude to load is absent from the bundle — and moderate redundancy between the Quick reference, Files, and Behavioural rules sections.
Suggestions
Ship the 11 missing `rules/*.md` files (or remove the references and inline the essential rules): Phase 2–4 instruct 'Load `rules/chart-type-selection.md`', 'Load `rules/anti-patterns.md`', etc., but no `rules/` directory exists in the bundle, so the core workflow cannot execute as written.
Add an explicit validation checkpoint at the end of the workflow (e.g. 'Before responding, verify the answer fills the Output format template and every Accessibility check cites a named anti-pattern or WCAG criterion') to close the workflow-clarity gap.
Trim duplication between the Quick reference tables, the Files section, and Behavioural rules — e.g. keep the intent→chart table only as a pointer to `rules/chart-type-selection.md`, since Behavioural rules 2/5/8 already restate Phase 4/5 guidance and the Hard rules.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence (no explanations of what charts or libraries are), and instruction lines like "Ask — in **one** batched message — only the answers you cannot infer from context" are tight. But there is duplicated material: the Quick reference tables restate content flagged as "full table in `rules/chart-type-selection.md`", the Files section re-describes files already linked in the Phase 4 table, and Behavioural rules 2/5/8 restate Phase 4/5 guidance and Hard rules. That repetition is what keeps it at level 4 ('minor instances of over-explanation that could be trimmed') rather than 5. | 4 / 5 |
Actionability | Guidance is fully concrete for an advisory skill: a copy-paste output-format template, a signal→file loading table, an intent→chart decision table, a platform→library table with named defaults ("shadcn charts (Recharts under the hood) or Tremor", "Apache ECharts (Canvas)", "Victory Native XL (Skia)"), and numeric hard rules ("Pie charts: ≤ 5 slices", "Categorical palette: ≤ 8 categories"). Per the rubric's instruction-only note, absence of code is not penalized when guidance is this specific — the common cases are covered concretely. | 5 / 5 |
Workflow Clarity | Six phases are clearly sequenced (discover intent → select chart → select library → apply guardrails → point at examples → hand off), with batching rules in Phase 1 ("If the user already supplied any of these, do not ask again") and a hard guard in Phase 2 ("Never recommend a chart whose intent does not match the user's question"). This is an advisory skill with no destructive/batch operations, so no validation cap applies; what keeps it at 4 rather than 5 is the absence of any explicit verification checkpoint (e.g., confirming the assembled answer matches the Output format before responding). | 4 / 5 |
Progressive Disclosure | The design is strong — a lean overview with clearly signaled one-level-deep references, an always-on vs load-on-demand split, and a signal table mapping request cues to exact files — which alone would suggest 5. But scored against the actual bundle as the rubric requires, 10 of the 11 referenced rule files (e.g. `rules/chart-type-selection.md`, `rules/anti-patterns.md`, `rules/performance.md`) do not exist; only `references/galleries-and-examples.md` is present. Phases 2–4 instruct loading files that are missing, so the disclosure structure is broken in delivery. Not level 2 because the inlined content (quick-reference tables) partially covers the missing detail and the references that do exist are clearly signaled and well organized. | 3 / 5 |
Total | 16 / 20 Passed |