Content
85%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 unusually dense, expert-level migration runbook with excellent sequencing, validation checkpoints, and copy-paste commands; nearly every token is directive. Its one real defect is that the entire detailed layer — prop tables, class mappings, worked examples — is delegated to reference files that are absent from this bundle, leaving the skill's core transformation knowledge unreachable.
Suggestions
Ship the referenced bundle files (universal-patterns.md, consumer-props.md, overlays.md, menus.md, form-controls.md, disclosure.md, display-misc.md, class-mapping.md, wrapper-shapes.md) in a references/ directory — the body delegates all concrete prop/class mapping tables to them, and they are currently missing, so the core transformation step cannot be executed from the bundle.
Since consumer call-site prop rewrites ('repoint consumers ONE AT A TIME... the call-site props in consumer-props.md') are the main break surface per the skill's own analysis, inline at least one worked example of a before/after consumer diff in SKILL.md so the most common case survives a missing or relocated reference file.
Add a short bundle manifest near the top (one line listing the reference files with their one-line purposes) so a reader can immediately verify the expected bundle is complete before starting a migration.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is relentlessly lean: no explanation of what Radix, React, shadcn, or migrations are — every line is directive ("diff the user's file against its stock origin, using the components.json style VERBATIM in the URL"). Even rationale is given only where it changes behavior ("A clean merge is NOT proof of a clean file"). Matches the 'every token earns its place' anchor. | 5 / 5 |
Actionability | Highly concrete and copy-paste ready: exact commands (`npx shadcn@latest info --json`, `git merge-file user.tsx radix-golden.tsx base-golden.tsx`, `grep -n "radix-ui\|@radix-ui"`), exact URL patterns, exact file paths and modes. Not 5 because the detailed mapping tables and worked examples (consumer prop rewrites, asChild->render example) are delegated to reference files (consumer-props.md, universal-patterns.md) that are not present in this bundle, so the common-case transformations are not executable from what ships here. | 4 / 5 |
Workflow Clarity | Clear multi-step sequences with explicit validation checkpoints throughout for a batch/destructive operation: baseline typecheck before touching dependencies (step 4 of Preflight), typecheck per repointed consumer, mandatory leftover grep sweep on every golden-pair file with the failure mode explained, build per batch and full build vs baseline at the end, plus resume-from-disk state handling and a fixed report structure. Feedback loops (validate -> fix -> re-validate) are explicit. | 5 / 5 |
Progressive Disclosure | The body itself is well structured with clearly signaled one-level-deep references (universal-patterns.md, overlays.md, menus.md, form-controls.md, disclosure.md, display-misc.md, class-mapping.md, wrapper-shapes.md, consumer-props.md, each named with its purpose). However, none of these files exist — the bundle contains no references/, scripts/, or assets/ directories at all — so navigation dead-ends and the detailed content is unreachable. Not 2 because the SKILL.md is not a wall of inlined reference text and the references are clearly signaled, not buried; not 4/5 because scoring against the actual bundle structure, the split is only nominal. | 3 / 5 |
Total | 17 / 20 Passed |