Content
56%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 SKILL.md body is a well-structured, prescriptive workflow with good sequencing, version gating, and an anti-hallucination checklist. Its main weakness is that the core generation path relies on example/fragment/bean/dependency/properties files that are not present in the bundle, leaving the actual code-generation step non-executable and the reference graph partially broken.
Suggestions
Bundle the missing example trees (examples/_skeletons, _fragments, _beans, _dependencies, _properties) or replace the "Read ... file" instructions with inline skeletons/fragments so the generation step is executable as written.
Deduplicate the context-first guidance: Step 0 restates the Decision-making principle (levels 1–3) almost verbatim — consolidate into a single canonical statement and cross-reference it.
Add a concrete, runnable validation step (e.g. a refresh_build_system_model output check or a compile/parse verification command) to close the validation loop rather than relying solely on the static anti-hallucination checklist.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and well-organized with tables and checklists, but the Defaults/decision-making/Step 0 sections repeat the same "derive from context, skip already-answered questions" guidance several times (e.g. Step 0 restates principles 1–3 verbatim), which could be tightened without losing clarity. | 3 / 5 |
Actionability | The instructions are concrete and prescriptive (named MCP tools, exact reference mappings, numbered generation steps), but the central generation step delegates to files the bundle does not contain — Step 4 says "Read skeleton from examples/_skeletons/{lang}.md" and fragments/beans/dependencies/properties, none of which exist, so Claude cannot actually execute the core code generation as written. | 3 / 5 |
Workflow Clarity | Clear Step 0–5 sequence with explicit checkpoints (existing-config collision warning, bootMajor version gating for Authorization Server, an anti-hallucination checklist, and "STOP and ask user" when no matching example exists); the only gap is that the validation loop is an instruction-level checklist rather than an executable verify command, since the example files needed to perform the generation are missing. | 4 / 5 |
Progressive Disclosure | The body points one level deep to real reference files (references/jwt.md, common-dsl.md, etc. all exist), but those references in turn point to a second level (_fragments/, _beans/, _dependencies/, _properties/, _skeletons/) that is not bundled, creating broken multi-level navigation; the guideline scores against actual bundle structure, and the referenced paths are largely absent. | 3 / 5 |
Total | 13 / 20 Passed |