Content
47%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill has excellent workflow structure and thorough verification checklists per recipe, but is severely undermined by verbosity. Inline essays on signals-based reactivity, state machines, branded types, and locality of behaviour belong in reference files, not the main SKILL.md. The complete absence of executable code examples is a significant gap for a frontend implementation skill — concrete component snippets would dramatically improve actionability.
Suggestions
Move the essay-length bullet points in Core Contract (signals convergence, state machines, locality of behaviour, branded types) to reference files and replace with 1-line summaries linking to those files.
Add at least 2-3 executable TypeScript/React code examples showing the patterns Artisan prescribes (e.g., a minimal component with error boundary, a form with RHF+Zod, a Server Component boundary).
Remove the HTML capabilities summary comment — it duplicates information already in the body and consumes ~40 lines of tokens that provide no runtime value.
Consolidate the repeated table structures (Framework Coverage, Cross-Framework Patterns, Reference Map) which overlap significantly in their reference pointers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Extremely verbose at ~300+ lines. Extensive explanations of concepts Claude already knows (what signals-based reactivity is, what branded types are, how React Compiler works). The capabilities summary HTML comment, lengthy 'Never' list with CVE details, cross-framework convergence essays, and repeated information across tables all waste significant tokens. Much content reads like a blog post rather than actionable instructions. | 1 / 3 |
Actionability | Provides concrete workflow phases, specific tool/library recommendations (Zod v4, TanStack Query v5, RHF), and detailed VERIFY checklists per recipe. However, there are zero executable code examples anywhere in the skill — no sample component, no code snippet showing the patterns it prescribes. The guidance is specific but not copy-paste ready. | 2 / 3 |
Workflow Clarity | The ANALYZE → DESIGN → IMPLEMENT → VERIFY → HANDOFF workflow is clearly sequenced with a table mapping phases to actions and required reads. Each recipe subcommand has explicit VERIFY gates that serve as validation checkpoints. The dispatch logic (parse first token, match to recipe, load specific references) is unambiguous. Error recovery is implicit but the verification steps are thorough. | 3 / 3 |
Progressive Disclosure | References to 9+ reference files are well-signaled with clear 'Read this when' guidance in the Reference Map table. However, no bundle files were provided, so we cannot verify these references exist. More critically, the SKILL.md itself is monolithic — the core contract section alone contains multiple essay-length bullet points (signals convergence, state machines, locality of behaviour, branded types) that should be in reference files rather than inline. | 2 / 3 |
Total | 8 / 12 Passed |